Spinal cord injury post-neural function comprehensive evaluation system based on big data

Abstract

The application discloses a spinal cord injury post-neural function comprehensive evaluation system based on big data, which comprises a data acquisition module, a spinal cord injury multi-modal feature depth selection module, a pre-injury neural function baseline construction module and a double-reference spinal cord injury neural function comprehensive evaluation module. The application relates to the technical field of medical data processing, in particular to a spinal cord injury post-neural function comprehensive evaluation system based on big data. The pre-injury neural function baseline construction module and the double-reference spinal cord injury neural function comprehensive evaluation module are innovatively combined, so that the overallness and accuracy of the evaluation result are improved. A collaborative correction feature screening algorithm fusing ternary mutual information is proposed, so that the model evaluation precision and stability are improved. The clustering algorithm is improved through the fusion of a high-dimensional vector included angle variance quantification method and a core point secondary evaluation strategy based on a reverse neighborhood quantity evaluation mechanism, so that the clustering result accuracy is improved, and the accuracy of the spinal cord injury post-neural function evaluation result is improved.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a comprehensive assessment system for neurological function after spinal cord injury based on big data. Background Technology

[0002] The Comprehensive Neurological Function Assessment System after Spinal Cord Injury is an intelligent medical and health information management system for the entire rehabilitation cycle of spinal cord injury patients. By collecting multi-source data from spinal cord injury patients throughout the entire cycle, and relying on medical big data processing and artificial intelligence information analysis technology, it conducts comprehensive information-based assessment and dynamic monitoring of the neurological function damage status and recovery progress of patients after spinal cord injury. It can provide medical management entities with dynamic tracking of neurological function recovery, accurate evaluation of rehabilitation efficacy, and clinical decision support, realizing intelligent information processing, precise assessment, and standardized management of neurological function rehabilitation throughout the entire cycle after spinal cord injury.

[0003] However, traditional comprehensive assessment systems for neurological function after spinal cord injury only use a single post-injury grading benchmark, resulting in a single dimension of comprehensive assessment and inaccurate assessment results. Traditional feature selection algorithms can only perform single feature weight evaluation and ignore the interaction effects between multimodal features, resulting in biased feature importance quantification, incomplete removal of redundant features, and underestimation of features with high synergistic value. Existing clustering algorithms applicable to post-injury neurological function assessment models have technical problems such as reliance on manual global parameter tuning and poor compatibility with high-dimensional multimodal data calculations, leading to inaccurate clustering results and insufficient accuracy of model output results. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a comprehensive assessment system for neurological function after spinal cord injury based on big data. Addressing the technical problem that traditional comprehensive assessment systems for neurological function after spinal cord injury only use a single post-injury group grading benchmark, resulting in a limited scope and inaccurate assessment results, this solution innovatively combines a pre-injury neurological function baseline construction module and a dual-benchmark comprehensive assessment module for spinal cord injury. This dual-benchmark comprehensive assessment system uses a pre-injury specific baseline as the first benchmark and the clustering results of a homogeneous baseline group as the second benchmark, simultaneously considering both individual longitudinal functional changes before and after injury and the horizontal hierarchical positioning of the homogeneous group. This invention effectively compensates for the inherent defects of traditional single assessment benchmarks, improves the comprehensiveness and accuracy of assessment results, enhances the repeatability of neurological function grading assessment after spinal cord injury, and achieves integrated comprehensive assessment of longitudinal individual changes in neurological function and horizontal group hierarchical positioning after spinal cord injury. Addressing the technical problems of traditional feature selection algorithms, which can only perform single feature weight assessment and ignore the interaction effects between multimodal features, resulting in biased feature importance quantification, incomplete removal of redundant features, and underestimation of high-co-value features, thus leading to insufficient accuracy of the assessment model, this invention innovatively proposes a collaborative correction feature selection algorithm that integrates ternary mutual information. This algorithm, by superimposing special features on the initial weights... The final feature weights are calculated using ternary mutual information between features, and an adaptive selection threshold is constructed based on these final feature weights to select a subset of core features. This effectively improves the accuracy and rationality of feature importance ranking, enhances the ability to mine the correlation and synergistic effects between multimodal features, and improves the filtering efficiency of redundant and noisy features. It achieves precise, adaptive, and highly discriminative selection of multimodal features in spinal cord injury scenarios, providing a streamlined and efficient feature input for subsequent neurofunctional assessment models, and improving the model's evaluation accuracy and stability. This addresses the shortcomings of existing clustering algorithms applicable to post-injury neurofunctional assessment models, which rely on manual global parameter tuning and have compatibility issues with high-dimensional multimodal data computation. This solution addresses the technical problem of inaccurate clustering results and insufficient precision in model output by integrating a high-dimensional vector angle variance quantification method with a core point secondary evaluation strategy based on the inverse neighborhood quantity evaluation mechanism. This completely eliminates subjective errors from manual parameter tuning, accurately adapts to data, is compatible with high-dimensional multimodal data, and significantly improves clustering efficiency and accuracy. This enhances the accuracy of the output results of the neurological function assessment model after spinal cord injury, providing a stable and repeatable homogeneous population horizontal comparison benchmark for the dual-benchmark comprehensive assessment system of spinal cord injury neurological function. Ultimately, it achieves a comprehensive, precise, and intelligent assessment of neurological function after spinal cord injury in patients.

[0005] The technical solution adopted by the present invention is as follows: The comprehensive assessment system for neurological function after spinal cord injury based on big data provided by the present invention includes a data acquisition module, a deep selection module for multimodal features of spinal cord injury, a baseline construction module for pre-injury neurological function, and a comprehensive assessment module for neurological function after spinal cord injury based on dual benchmarks.

[0006] The data acquisition module obtains optimized neural function assessment data through data collection and data preprocessing operations.

[0007] The spinal cord injury multimodal feature depth selection module specifically calculates the ternary mutual information between features, calculates the final feature weight by superimposing the ternary mutual information between features on the initial weight, and constructs an adaptive screening threshold based on the final feature weight to screen the core feature subset, thereby completing the construction of the collaborative correction feature screening algorithm. Then, the reference patient pre-injury assessment data and reference patient post-injury assessment data in the neurological function assessment optimization data are respectively input into the feature screening algorithm, and the output is the core feature subset of pre-injury neurological function assessment and the core feature subset of post-injury neurological function assessment.

[0008] The pre-injury neurological function baseline construction module specifically uses a fully connected neural network to construct a pre-injury neurological function assessment model, trains the model, and finally inputs the target pre-injury neurological function assessment core feature data into the trained model to obtain the target patient's pre-injury neurological function baseline results.

[0009] The dual-benchmark spinal cord injury neurological function comprehensive assessment module specifically improves the clustering algorithm by fusing a high-dimensional vector angle variance quantification method with a core point secondary assessment strategy based on the inverse neighborhood quantity assessment mechanism. The improved clustering algorithm is used as the post-injury neurological function assessment model. Then, based on the target patient's pre-injury neurological function baseline results as the screening benchmark, homogeneous screening of the pre-injury baseline is performed to obtain the input data for the post-injury neurological function assessment model. This data is then input into the post-injury neurological function assessment model to obtain the target patient's post-spinal cord injury neurological function assessment results. Finally, based on the target patient's pre-injury neurological function baseline results and the post-spinal cord injury neurological function assessment results, a comprehensive assessment of post-spinal cord injury neurological function is achieved.

[0010] Furthermore, the data acquisition module specifically obtains raw data for neurological function assessment by performing data acquisition operations, and performs data preprocessing on the raw data for neurological function assessment to obtain optimized data for neurological function assessment.

[0011] The raw data for neurological function assessment includes pre-injury assessment data of the reference patient, pre-injury assessment data of the target patient, post-injury assessment data of the reference patient, and post-injury assessment data of the target patient.

[0012] The data preprocessing specifically involves sequentially performing data cleaning and normalization on the raw data of the neurological function assessment to obtain optimized data for the neurological function assessment.

[0013] Furthermore, the spinal cord injury multimodal feature depth selection module specifically includes the following steps:

[0014] The collaborative correction feature selection algorithm is constructed by including the following steps:

[0015] To construct a feature screening sample matrix, the reference patient injury assessment data in the neurological function assessment optimization data is divided into a multimodal feature set and a sample standard label set. Based on the multimodal feature set and the sample standard label set, a feature screening sample matrix is ​​constructed.

[0016] The initial weight calculation for a single feature is specifically based on the ReliefF feature selection algorithm, which calculates the initial weights of the features in the feature selection sample matrix to obtain the initial weights of all features.

[0017] The calculation of ternary mutual information between features involves first discretizing the continuous features in the multimodal feature set using equal-frequency methods, dividing them into M equal-frequency intervals to convert the continuous features into discrete features. Then, for any two different features in the multimodal feature set, the mutual information of single features and the joint mutual information of feature pairs are calculated sequentially using the standard label set of the samples. Finally, the ternary mutual information value is calculated for each feature pair, and this process is repeated for all feature pairs in the multimodal feature set to complete the ternary mutual information calculation. The formula used is as follows:

[0018] ;

[0019] In the formula, and Let each be represented as any two distinct features in the multimodal feature set. This represents the mutual information between the i-th feature and the label set C. Representation of features and The joint mutual information between the feature pairs formed and the label set C. Representation of features , The ternary mutual information value between the label set C and the label set C. This represents the mutual information between the j-th feature and the label set C;

[0020] The final weight calculation of features is specifically based on the ternary mutual information value between features, which corrects the initial weights of the features to obtain the final weights of all features.

[0021] The core feature subset selection process involves using the average final weight of all features as the adaptive selection threshold, retaining features with a final weight greater than the adaptive selection threshold, removing features with a weight less than or equal to the adaptive selection threshold, and combining all features that meet the selection rules to obtain the core feature subset.

[0022] The pre-injury neurological function assessment feature screening specifically involves inputting the reference patient's pre-injury assessment data from the optimized neurological function assessment data into a collaborative correction feature screening algorithm to obtain a subset of core features for pre-injury neurological function assessment.

[0023] The feature selection for post-injury neurological function assessment involves inputting the reference patient post-injury assessment data from the optimized neurological function assessment data into a collaborative correction feature selection algorithm to obtain a subset of core features for post-injury neurological function assessment.

[0024] Furthermore, the pre-injury neural function baseline construction module specifically includes the following steps:

[0025] The pre-injury neurological function assessment model is constructed and trained. Specifically, a fully connected neural network is used to build the pre-injury neurological function assessment model. Then, the core feature data of the reference pre-injury neurological function assessment is input into the pre-injury neurological function assessment model for model training, and the trained pre-injury neurological function assessment model is obtained.

[0026] To obtain the baseline of pre-injury neurological function of the target patient, the core feature data of the pre-injury neurological function assessment are input into the trained pre-injury neurological function assessment model to obtain the pre-injury neurological function assessment results of the target patient, which serve as the baseline results of the pre-injury neurological function of the target patient.

[0027] Furthermore, the dual-benchmark spinal cord injury neurological function comprehensive assessment module specifically includes the following steps:

[0028] Constructing a post-injury neurological function assessment model involves establishing an improved clustering algorithm to build the model, including the following steps:

[0029] The adaptive orientation centrality metric is calculated by employing a high-dimensional vector angle variance quantization method. For each data point, the Euclidean distance between it and other data points is calculated. All other data points are then sorted in ascending order of their distance values, and the k nearest neighbors with the smallest distances at the top of the sort are selected to form the k-nearest neighbor set for that data point. Then, using that data point as the starting point, high-dimensional space vectors are drawn to any two different data points in the k-nearest neighbor set, and the angle between the two vectors is calculated. This process is repeated for all pairwise distinct point combinations in the k-nearest neighbor set to complete the angle calculation. Finally, the statistical variance of all pairwise angles is calculated to obtain the adaptive orientation centrality metric value for that data point.

[0030] The population average directional centrality metric is calculated by taking the adaptive directional centrality metric value of each data point as the basis. For each data point, the average of the adaptive directional centrality metric values ​​of all data points in the k nearest neighbor set of that data point is calculated to obtain the population average directional centrality metric of that data point.

[0031] The data points are initially divided adaptively. Specifically, all data points are traversed, and the adaptive directional centrality metric of each point is compared with the average directional centrality metric of the population. If the adaptive directional centrality metric of the data point is less than the average directional centrality metric of the population, the data point is assigned to the core point set; otherwise, the data point is assigned to the boundary point set. Finally, the internal point set and the boundary point set are obtained.

[0032] The core point secondary evaluation is specifically based on the reverse neighborhood quantity evaluation mechanism. First, for each data point in the internal point set, the reverse k-nearest neighbor set of that data point is calculated. Then, the number of data points in the reverse k-nearest neighbor set of that data point is counted. If the data point If k is true, then the data point is determined to be a core point; otherwise, the data point is a pseudo-core point.

[0033] The inverse k-nearest neighbor set specifically refers to the set of internal points. For the target data point, iterate through all internal points and select all internal points j that include the target data point in their own k-nearest neighbor set. The selected internal points form the inverse k-nearest neighbor set of the target data point.

[0034] The core cluster expansion process involves first marking all internal points as unassigned, using each unassigned core point as the initial cluster center, then merging the k-nearest neighbor set and the reverse k-nearest neighbor set of the core point to form an expanded neighborhood. Subsequently, a double constraint check is performed on all unassigned internal points in the expanded neighborhood, and only points that simultaneously satisfy both conditions can be assigned to the cluster of the current core point. Finally, the newly added data point in the cluster is used as the new cluster center, and the expansion neighborhood and double constraint check are iteratively executed until no new internal data points are added to the current cluster. Finally, all core points are traversed to complete the cluster partitioning of all internal points.

[0035] The dual constraint condition verification specifically includes orientation centrality consistency constraint and reverse neighborhood density constraint;

[0036] The directional centrality consistency constraint specifically means that the adaptive directional centrality measure of the internal point to be assigned and the core point that serves as the cluster center are less than the group average directional centrality measure of the two.

[0037] The reverse neighborhood density constraint requires that the number of data points in the reverse k-nearest neighbor set of the internal points to be assigned be greater than or equal to the minimum neighborhood threshold. ;

[0038] The remaining point allocation process involves searching for the nearest allocated internal point for each data point in the boundary point set, assigning the cluster label of that internal point to the current boundary point, and then searching for the nearest allocated data point for the remaining unassigned data points, assigning the cluster label of that point to the current unassigned data point.

[0039] The clustering results are output by merging all data points into the corresponding cluster set after all data points have been assigned, forming multiple independent clusters. Each cluster represents a category of the neurological function assessment results after a patient's spinal cord injury.

[0040] The pre-injury baseline homogeneity screening is specifically carried out by using the pre-injury neurological function baseline results of the target patient as the screening criterion. All reference patients are screened to select those whose pre-injury neurological function assessment results are the same as those of the target patient. The post-injury assessment data of the selected reference patients are then extracted. The post-injury assessment data of the selected reference patients and the post-injury assessment data of the target patient are then subjected to feature extraction according to the core feature subset of post-injury neurological function assessment to obtain feature data, which is used as the input data of the post-injury neurological function assessment model.

[0041] Real-time assessment of neurological function after spinal cord injury involves inputting the data from the neurological function assessment model into the model, generating real-time clustering output results, and statistically analyzing the cluster labels in each cluster based on the real-time clustering output results. The cluster label with the highest frequency of occurrence is selected as the neurological function assessment result label for that cluster, thus obtaining the neurological function assessment result of the target patient after spinal cord injury.

[0042] Comprehensive assessment of neurological function after spinal cord injury involves comprehensively evaluating the neurological function of the target patient after spinal cord injury based on the target patient's baseline neurological function before injury and the neurological function assessment results after spinal cord injury.

[0043] The beneficial effects achieved by the present invention using the above solution are as follows:

[0044] (1) In response to the technical problem that traditional comprehensive assessment systems for neurological function after spinal cord injury only use a single post-injury group grading benchmark, resulting in a single dimension of comprehensive assessment and inaccurate assessment results, this solution innovatively combines a pre-injury neurological function baseline construction module and a dual-benchmark comprehensive assessment module for spinal cord injury. The dual-benchmark comprehensive assessment system uses the pre-injury exclusive baseline as the first benchmark and the results of clustering of homogeneous groups based on the baseline as the second benchmark. It also takes into account the longitudinal functional changes of individuals before and after injury and the horizontal hierarchical positioning of homogeneous groups. This effectively makes up for the inherent defects of traditional single assessment benchmarks, improves the comprehensiveness and accuracy of assessment results, enhances the repeatability of neurological function grading assessment after spinal cord injury, and realizes the integrated comprehensive assessment of the quantification of longitudinal individual changes in neurological function after spinal cord injury and the horizontal hierarchical positioning of groups.

[0045] (2) To address the technical problems of traditional feature selection algorithms, which can only evaluate single feature weights and ignore the interaction effects between multimodal features, resulting in biased feature importance quantification, incomplete removal of redundant features, and underestimation of high synergistic value features, thus leading to insufficient accuracy of the evaluation model, this invention innovatively proposes a collaborative correction feature selection algorithm that integrates ternary mutual information. By superimposing ternary mutual information between features on the initial weights to calculate the final feature weights, and constructing an adaptive screening threshold based on the final feature weights to screen the core feature subset, this invention effectively improves the accuracy and rationality of feature importance ranking, enhances the ability to mine the correlation and synergistic effects between multimodal features, improves the filtering efficiency of redundant and noisy features, and realizes the precise, adaptive, and highly discriminative selection of multimodal features in the spinal cord injury scenario. This provides a simplified and efficient feature input for the subsequent neural function evaluation model, and improves the model evaluation accuracy and stability.

[0046] (3) In view of the technical problems of existing clustering algorithms applicable to post-injury neurological function assessment models, such as reliance on manual global parameter tuning and poor compatibility with high-dimensional multimodal data, resulting in inaccurate clustering results and insufficient accuracy of model output results, this solution innovatively improves the clustering algorithm by integrating the high-dimensional vector angle variance quantification method with the core point secondary assessment strategy based on the reverse neighborhood quantity assessment mechanism. This completely eliminates subjective errors of manual parameter tuning, accurately adapts data, is compatible with high-dimensional multimodal data, and significantly improves clustering efficiency and accuracy of clustering results. This improves the accuracy of the output results of the post-spinal cord injury neurological function assessment model and provides a stable and repeatable homogeneous population horizontal comparison benchmark for the dual-benchmark spinal cord injury neurological function comprehensive assessment system. Ultimately, it realizes a comprehensive, precise, and intelligent assessment of the neurological function of patients after spinal cord injury. Attached Figure Description

[0047] Figure 1A schematic diagram of the modules of the comprehensive assessment system for neurological function after spinal cord injury based on big data provided by the present invention;

[0048] Figure 2 A flowchart illustrating the multimodal feature depth selection module for spinal cord injury;

[0049] Figure 3 A flowchart illustrating the construction of a collaborative correction feature selection algorithm in the multimodal feature depth selection module for spinal cord injury;

[0050] Figure 4 A flowchart illustrating the comprehensive assessment of neurological function in dual-benchmark spinal cord injury.

[0051] Figure 5 This is a flowchart illustrating the process of constructing a post-injury neurological function assessment model within the dual-benchmark spinal cord injury neurological function comprehensive assessment module.

[0052] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0053] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0054] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0055] Example 1, see Figure 1 The present invention provides a comprehensive assessment system for neurological function after spinal cord injury based on big data, including a data acquisition module, a deep selection module for multimodal features of spinal cord injury, a baseline construction module for pre-injury neurological function, and a comprehensive assessment module for neurological function after spinal cord injury based on dual benchmarks.

[0056] The data acquisition module obtains optimized neurological function assessment data through data acquisition and data preprocessing operations, and sends the data to the spinal cord injury multimodal feature depth selection module, the pre-injury neurological function baseline construction module, and the dual-benchmark spinal cord injury neurological function comprehensive assessment module.

[0057] The spinal cord injury multimodal feature depth selection module receives data sent by the data acquisition module. Specifically, it calculates the ternary mutual information between features, calculates the final feature weight by superimposing the ternary mutual information between features on the initial weight, and constructs an adaptive screening threshold based on the final feature weight to screen the core feature subset, thereby completing the construction of the collaborative correction feature screening algorithm. Then, the reference patient pre-injury assessment data and reference patient post-injury assessment data in the neurological function assessment optimization data are respectively input into the feature screening algorithm, and the output is the core feature subset of pre-injury neurological function assessment and the core feature subset of post-injury neurological function assessment. The data is then sent to the pre-injury neurological function baseline construction module and the dual-benchmark spinal cord injury neurological function comprehensive assessment module.

[0058] The pre-injury neurological function baseline construction module receives data sent by the data acquisition module and the spinal cord injury multimodal feature depth selection module. Specifically, it constructs a pre-injury neurological function assessment model using a fully connected neural network, trains the model, and finally inputs the target pre-injury neurological function assessment core feature data into the trained model to obtain the target patient's pre-injury neurological function baseline result.

[0059] The dual-benchmark spinal cord injury neurological function comprehensive assessment module receives data from the data acquisition module, the spinal cord injury multimodal feature depth selection module, and the pre-injury neurological function baseline construction module. Specifically, it improves the clustering algorithm by fusing a high-dimensional vector angle variance quantification method with a core point secondary assessment strategy based on the inverse neighborhood quantity assessment mechanism. The improved clustering algorithm is used as the post-injury neurological function assessment model. Then, based on the pre-injury neurological function baseline results of the target patient, homogeneity screening of the pre-injury baseline is performed to obtain the input data of the post-injury neurological function assessment model. This data is then input into the post-injury neurological function assessment model to obtain the post-spinal cord injury neurological function assessment results of the target patient. Finally, based on the pre-injury neurological function baseline results and the post-spinal cord injury neurological function assessment results of the target patient, a comprehensive assessment of post-spinal cord injury neurological function is achieved.

[0060] By performing the above operations, this solution addresses the technical problem in traditional comprehensive assessment systems for post-spinal cord injury neurological function that rely solely on a single post-injury group grading benchmark, resulting in a limited scope and inaccurate assessment results. It innovatively combines a pre-injury neurological function baseline construction module and a dual-benchmark comprehensive assessment module for spinal cord injury neurological function. This constructs a dual-benchmark comprehensive assessment system with a pre-injury baseline as the first benchmark and the results of homogeneous group clustering as the second benchmark. Simultaneously, it considers both individual longitudinal functional changes before and after injury and the horizontal hierarchical positioning of homogeneous groups. This effectively compensates for the inherent defects of traditional single-benchmark assessments, improves the comprehensiveness and accuracy of assessment results, enhances the repeatability of post-spinal cord injury neurological function grading assessments, and achieves an integrated comprehensive assessment of both the quantification of longitudinal individual changes in spinal cord injury neurological function and the horizontal hierarchical positioning of groups.

[0061] Example 2, see Figure 1 This embodiment is based on the above embodiment. Specifically, the data acquisition module performs data acquisition operations through patient responses and examinations to obtain raw data for neurological function assessment, and performs data preprocessing on the raw data for neurological function assessment to obtain optimized data for neurological function assessment.

[0062] The raw data for neurological function assessment includes pre-injury assessment data of the reference patient, pre-injury assessment data of the target patient, post-injury assessment data of the reference patient, and post-injury assessment data of the target patient.

[0063] The reference patient pre-injury assessment data and the target patient pre-injury assessment data both include the patient's basic physiological data, the patient's occupational work behavior data, the patient's daily activity behavior data, and the patient's health data;

[0064] The patient's basic physiological data includes age, gender, height, weight, BMI, age group stratification, and congenital limb development.

[0065] The patient's occupational work behavior data includes occupation type, years of employment, daily labor intensity level, occupational limb use habits, and special occupational circumstances;

[0066] The patient's daily activity behavior data includes daily exercise frequency, exercise type, years of exercise, average daily activity level, and average daily sedentary time;

[0067] The patient's health data includes information on chronic underlying diseases, years since diagnosis, daily control status, history of non-invasive surgery, history of non-neurological major diseases, status of daily living self-care ability before injury, smoking history, drinking history, and long-term regular medication before injury.

[0068] The reference patient pre-injury assessment data also includes the reference patient pre-injury neurological function assessment results;

[0069] The reference patient's pre-injury neurological function assessment results are divided into normal neurological function, mild neurological function, moderate neurological function, and severe neurological function.

[0070] The reference patient post-injury assessment data and the target patient post-injury assessment data both include patient basic physiological data, patient occupational work behavior data, patient rehabilitation behavior data, patient health data, spinal cord injury data, medical structure information data, and neurological function scale data;

[0071] The patient rehabilitation behavior data includes gait parameters, balance parameters, muscle activity frequency data, sleep data, heart rate variability data, postural change data, daily activity intensity data, urination and defecation behavior data, rehabilitation training execution frequency and duration data, and rehabilitation training movement completion data.

[0072] The spinal cord injury data includes the cause of injury, precise vertebral segment of injury, level of spinal cord injury, extent of injury, type of tract involved, nature of injury, grade of injury severity, pathological type of injury, stage of disease, stable condition, and surgical treatment information.

[0073] The medical structural information data includes complete blood biochemical indicators, metabolic-related indicators, immune-related indicators, spinal cord lesion length, degree of spinal canal compression, range of intramedullary hemorrhage and edema, white matter fiber bundle continuity parameters, gray matter structural damage, and spinal bone structural damage.

[0074] The neurological function scale data include ASIA full score, Frankel grade, FIM functional independence score, VAS pain score, spasticity grade score, and sensory function specific score.

[0075] The reference patient post-injury assessment data also includes the results of neurological function assessment after spinal cord injury, and this is used as a cluster label;

[0076] The neurological function assessment results after spinal cord injury include no neurological deficit, mild neurological deficit, moderate neurological deficit, and severe neurological deficit.

[0077] The state of no neurological deficit specifically means that there is no significant difference in neurological function after the injury compared with the neurological function before the injury, and there is no functional decline that can be attributed to this spinal cord injury.

[0078] The term "mild neurological deficit" specifically refers to a clear and mild decline in neurological function after the injury compared to before the injury, with mild neurological deficit attributable to the current spinal cord injury. Core motor and sensory functions below the level of spinal cord injury are mostly preserved, and daily activities can be basically completed independently.

[0079] The aforementioned moderate neurological deficit state specifically refers to a significant moderate decline in neurological function after the injury compared to before the injury, with a moderate neurological deficit attributable to this spinal cord injury, while core motor and sensory functions below the level of spinal cord injury are partially preserved, and daily activities require assistance from others.

[0080] The severe neurological deficit state is specifically defined as a significant and substantial decline in neurological function after the injury compared to before the injury, with severe and complete neurological deficit attributable to this spinal cord injury, complete loss of motor and sensory functions below the level of spinal cord injury, and complete dependence on others for daily activities.

[0081] The data preprocessing is used to eliminate noise interference, systematic errors, data integrity defects and dimensional differences in the original data of neurofunctional assessment. Specifically, data cleaning and normalization are performed on the original data of neurofunctional assessment in sequence to obtain optimized data of neurofunctional assessment.

[0082] The data cleaning process is used to remove invalid and interfering data from the original data and to complete the data integrity. Specifically, it involves imputing missing values ​​and removing outliers from the original data. The imputation of missing values ​​is specifically done by using the mean imputation method to fill in missing values. The removal of outliers is specifically done by using the interquartile range method to detect and remove extreme values ​​in the original data.

[0083] The normalization process is used to eliminate the differences in dimensionality, numerical scale, and weight imbalance among different types of data. Specifically, it first uses a label encoding method to map the category fields in the original data to corresponding integer values ​​according to the category values, and then uses the Z-Score standardization method to standardize all continuous variables so that each continuous variable meets a uniform numerical scale range.

[0084] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment, based on the above embodiment, uses a multimodal feature depth selection module for spinal cord injury to adapt to the dual core task requirements of pre-injury neurological function baseline construction and post-injury dual-benchmark comprehensive neurological function assessment. It performs deep screening and dimensionality reduction on all multidimensional features in the optimized neurological function assessment data, eliminating redundant and noisy features unrelated to spinal cord injury neurological function assessment, and maximizing the retention of core features with strong correlation and strong clinical synergistic effects with neurological function status. Specifically, it includes the following steps:

[0085] A collaborative correction feature selection algorithm is constructed to complete feature importance quantification and core feature selection, specifically including the following steps:

[0086] A feature selection sample matrix is ​​constructed to clarify the feature inputs and the gold standard for neurological function assessment, providing a standardized input basis for subsequent feature weight calculations and adapting to the continuous and categorical heterogeneous characteristics of multimodal spinal cord injury data. Specifically, the reference patient injury assessment data in the optimized neurological function assessment data is divided into a multimodal feature set and a sample standard label set, and a feature selection sample matrix is ​​constructed based on the multimodal feature set and the sample standard label set. ,in, Let represent the d-dimensional feature vector of the first sample, i.e., the multimodal feature set of the first sample. Let represent the d-dimensional feature vector of the second sample, that is, the multimodal feature set of the second sample. This represents the d-dimensional feature vector of the nth sample, i.e., the multimodal feature set of the nth sample;

[0087] The multimodal feature set Where d represents the total dimension of the features, and each feature dimension corresponds to a quantitative indicator in the reference patient impairment assessment data within the neurological function assessment optimization data. Represents the d-th feature. Indicates the first feature, This indicates the second feature;

[0088] The sample standard label set Where n is the total number of patient samples, This is the label for the neurological function assessment result corresponding to the nth sample. This indicates the label representing the neurological function assessment result corresponding to the first sample. This indicates the neurological function assessment result label corresponding to the second sample; the neurological function assessment result label is divided into reference patient's pre-injury neurological function assessment result and reference patient's post-injury neurological function assessment result.

[0089] The reference patient injury assessment data includes reference patient pre-injury assessment data and reference patient post-injury assessment data;

[0090] The initial weight calculation for single features is used to initially quantify the basic discriminative ability of each single feature in the neurological functional grading of spinal cord injury, and to screen out features with independent discriminative power for neurological functional status. This solves the technical problems of computational redundancy and interference of evaluation results caused by the indiscriminate introduction of all features. Specifically, based on the ReliefF feature selection algorithm, the initial weights of the features in the feature screening sample matrix are calculated to obtain the initial weights of all features.

[0091] The ReliefF feature selection algorithm specifically involves setting the number of iterations to the total number of samples in the feature selection sample matrix, setting the nearest neighbor number K to 5, resetting the initial weights of all features to 0, and then randomly selecting one target sample from the feature selection sample matrix each time. The corresponding standard label for the sample is Class( ), find the target sample For samples with the same neurological function assessment result label, find the K nearest neighbors to construct a set of neighbors of the same type. At the same time, find the 5 nearest neighbors of the target sample in each class of samples with different labels to construct a set of neighbors of different types. Based on the set of neighbors of the same type and the set of neighbors of different types, update the weights of the features according to the ReliefF algorithm rules until all preset number of iterations are completed, and then output the initial weights of all features.

[0092] This study calculates ternary mutual information among features to uncover synergistic and redundant interaction effects among multimodal features in spinal cord injury. It addresses the core deficiency of the traditional ReliefF algorithm, which only evaluates single features and ignores pathological associations and combined effects between features, by accurately quantifying the information gain of two combined features for neurological function assessment. Specifically, the continuous features in the multimodal feature set are first discretized using equal-frequency methods, divided into M equal-frequency intervals, transforming continuous features into discrete features. Then, for any two different features in the multimodal feature set, single-feature mutual information and feature-pair joint mutual information are calculated sequentially using a standard label set. Finally, ternary mutual information values ​​are calculated for each feature pair, and this process is repeated for all feature pairs in the multimodal feature set to complete the ternary mutual information calculation. The formula used is as follows:

[0093] ;

[0094] ;

[0095] ;

[0096] In the formula, and Let each be represented as any two distinct features in the multimodal feature set. This represents the joint probability of the i-th feature and the c-th class of neurological function assessment result label, which is calculated using frequency statistics. This represents the marginal probability of the i-th feature. Representation of features and The joint marginal probability, This represents the marginal probability of the label for the class c neurological function assessment result. This represents the mutual information between the i-th feature and the label set C. Representation of features and The joint mutual information between the feature pairs formed and the label set C. Representation of features , The ternary mutual information value between the label set C and the label set C is used to quantify the cooperative interaction effect between the two features. When, it indicates a characteristic. and There is a synergistic effect; combining the two can provide more effective information for neural function assessment than the sum of single features. Otherwise, it indicates that the features... and There is information redundancy; combining the two methods cannot provide additional effective information for neurological function assessment. and Representing features respectively , Discretized values, Let represent the joint probability of the i-th feature and the j-th feature with label c. This represents the mutual information between the j-th feature and the label set C;

[0097] The final feature weight calculation is used to integrate the basic discriminative power of single features with the clinical synergistic effect of multiple features, resulting in feature importance weights that better align with the logic of neurological function assessment. This avoids the underestimation of synergistic feature weights and the overestimation of redundant feature weights in traditional methods. Specifically, the initial feature weights are corrected based on the ternary mutual information values ​​between features to obtain the final weights of all features. The formula used is as follows:

[0098] ;

[0099] In the formula, This represents the final weight of the i-th feature. Indicates the initial weight of the i-th feature;

[0100] The core feature subset selection is used to automatically select features with core value for neural function assessment, complete feature dimensionality reduction, and remove redundant, invalid, and noisy features. Specifically, the average of the final weights of all features is used as the adaptive selection threshold. Features with final weights greater than the adaptive selection threshold are retained, while features with weights less than or equal to the adaptive selection threshold are removed. All features that meet the selection rules are combined to obtain the core feature subset.

[0101] The pre-injury neurological function assessment feature screening specifically involves inputting the reference patient's pre-injury assessment data from the optimized neurological function assessment data into a collaborative correction feature screening algorithm to obtain a subset of core features for pre-injury neurological function assessment.

[0102] The core feature subset is a set of selected feature dimensions, without specific numerical values;

[0103] The feature selection for post-injury neurological function assessment involves inputting the reference patient post-injury assessment data from the optimized neurological function assessment data into a collaborative correction feature selection algorithm to obtain a subset of core features for post-injury neurological function assessment.

[0104] By performing the above operations, this invention innovatively proposes a collaborative correction feature selection algorithm that integrates ternary mutual information. This algorithm addresses the technical problems of traditional feature selection algorithms, which can only evaluate single feature weights and ignore the interaction effects between multimodal features, resulting in biased feature importance quantification, incomplete removal of redundant features, and underestimation of high-co-value features, ultimately leading to insufficient accuracy of the evaluation model. By superimposing ternary mutual information between features on the initial weights to calculate the final feature weights, and constructing an adaptive screening threshold based on the final feature weights to screen the core feature subset, this algorithm effectively improves the accuracy and rationality of feature importance ranking, enhances the ability to mine the correlation and synergistic effects between multimodal features, and improves the filtering efficiency of redundant and noisy features. It achieves accurate, adaptive, and highly discriminative selection of multimodal features in the context of spinal cord injury, providing a simplified and efficient feature input for subsequent neural function evaluation models and improving the model's evaluation accuracy and stability.

[0105] Example 4, see Figure 1 This embodiment is based on the above embodiment. The pre-injury neurological function baseline construction module is used to generate a pre-injury neurological function baseline result specific to the target patient based on the target patient's inherent characteristics before the injury, providing a unique personal control benchmark for subsequent post-injury neurological function deficit assessment. Specifically, it includes the following steps:

[0106] The pre-injury neurological function assessment model is constructed and trained. Specifically, a fully connected neural network is used to build the pre-injury neurological function assessment model. Then, the core feature data of the reference pre-injury neurological function assessment is input into the pre-injury neurological function assessment model for model training, and the trained pre-injury neurological function assessment model is obtained.

[0107] The pre-injury neurological function assessment model includes an input layer, two fully connected hidden layers, and an output layer. The input layer receives the core feature data of the pre-injury neurological function assessment, the fully connected hidden layers achieve non-linear feature mapping through the ReLU activation function, and the output layer uses the Softmax activation function to output the probability distribution of the neurological function classification, thus obtaining the patient's pre-injury neurological function assessment results.

[0108] The reference pre-injury neurological function assessment core feature data is the reference patient pre-injury assessment data in the neurological function assessment optimization data, and the feature data is obtained after feature extraction according to the core feature subset of pre-injury neurological function assessment.

[0109] The model training uses a multi-class cross-entropy loss function, and iteratively updates the weight matrix and bias parameters of the evaluation model through backpropagation algorithm and gradient descent optimization method. The evaluation model parameters are continuously optimized through multiple rounds of iteration. When the preset maximum number of training times is reached or the multi-class cross-entropy loss function value converges to a set threshold, the iterative training stops.

[0110] To obtain the baseline of pre-injury neurological function of the target patient, specifically, the core feature data of the pre-injury neurological function assessment of the target patient are input into the trained pre-injury neurological function assessment model to obtain the pre-injury neurological function assessment results of the target patient, which serve as the baseline results of the pre-injury neurological function of the target patient.

[0111] The target pre-injury neurological function assessment core feature data is the feature data obtained by extracting features from the target patient's pre-injury assessment data in the neurological function assessment optimization data according to the core feature subset of pre-injury neurological function assessment.

[0112] Example 5, see Figure 1 , Figure 4 and Figure 5 This embodiment, based on the above embodiment, uses the dual-benchmark spinal cord injury neurological function comprehensive assessment module as a unique individual control benchmark, combining the target patient's pre-injury neurological function baseline with the results of cluster analysis of the baseline homogeneous population, to accurately grade and comprehensively assess the degree of neurological function deficit after spinal cord injury; specifically, it includes the following steps:

[0113] A post-injury neurological function assessment model is constructed to achieve full-process automation, precision, and intelligence in neurological function impairment classification. Specifically, this involves establishing an improved clustering algorithm through the following steps: calculating adaptive directional centrality, calculating the average directional centrality of the population, adaptive initial data point partitioning, secondary evaluation of core points, expansion of core clusters, allocation of remaining points, and output of clustering results.

[0114] An adaptive orientation centrality metric is calculated to quantify the uniformity of local neighborhood orientation distribution of core feature data points in post-traumatic neurological function assessment, accurately reflecting the individual heterogeneity of neurological function impairment. Specifically, a high-dimensional vector angle variance quantification method is used. For each data point, its Euclidean distance to other data points is calculated. All other data points are sorted in ascending order of distance value. The k nearest neighbors of the data point with the smallest distance at the top of the sort are selected. Then, the data point is used as the starting point of a vector, and high-dimensional space vectors are drawn to any two different data points in the k nearest neighbors set. The angle between the two vectors is then calculated. This process is repeated for all pairwise different point combinations in the k nearest neighbors set. Finally, the statistical variance of all pairwise angles is calculated to obtain the adaptive orientation centrality metric value of the data point.

[0115] Specifically, the data points are the post-injury assessment data of the reference patient and the post-injury assessment data of the target patient in the optimized neurological function assessment data. These data points are feature data obtained after feature extraction according to a subset of the core features of post-injury neurological function assessment. Each data point is a core feature vector of post-injury neurological function assessment for a single patient. The formula used is as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] In the formula, This represents the path from data point i to its k nearest neighbor p. 3D space vector, This represents the path from data point i to its k nearest neighbor q. 3D space vector, This represents the number of feature dimensions of the core feature subset for post-injury neurological function assessment, i.e., the dimension of the feature vector for a single patient. Let represent the quantized value of the t-th feature of the nearest neighbor point p. Let represent the quantized value of the t-th feature of the nearest neighbor q. Let represent the quantized value of the t-th feature of data point i. Representing vectors and The angle between them This represents the function for calculating statistical variance. Let k represent the set of k nearest neighbors of data point i. The value represents the adaptive orientation centrality metric of data point i. The smaller the value, the more uniform the directional distribution of the patient's neurological function characteristics in the local neighborhood, and the more typical the case belongs to a certain neurological function impairment category. The larger the value, the more uneven the directional distribution, and the more typical the case belongs to the boundary transition between different impairment categories.

[0122] The population average directional centrality metric is calculated to provide a local reference for adaptive data point partitioning based on the overall distribution characteristics of local neural function population samples. This completely eliminates the reliance on manual global parameters in traditional clustering algorithms and avoids the global threshold failure problem caused by uneven distribution of neural function data. It can adapt to local density differences among different neural function subtypes. Specifically, based on the adaptive directional centrality metric value of each data point, the average of the adaptive directional centrality metric values ​​of all data points in the k-nearest neighbor set of that data point is calculated to obtain the population average directional centrality metric for that data point. The formula used is as follows:

[0123] ;

[0124] In the formula, This represents the population mean directional centrality measure of data point i. This represents the adaptive orientation centrality metric for data point j. Indicates the number of nearest neighbors;

[0125] The data point adaptive preliminary segmentation is used to adaptively distinguish core assessment data points from ordinary assessment data points of neurological function, and to achieve pre-extraction of the boundary contour of neurological function assessment. No manual setting of any segmentation threshold is required. The segmentation is completed by comparing the adaptive directional centrality of the sample itself with the average directional centrality of the local population. It is suitable for neurological function assessment scenarios of different age groups, different injury types, and different disease stages. Specifically, it traverses all data points and compares the adaptive directional centrality of each point with the average directional centrality of the population. If the adaptive directional centrality of the data point is less than the average directional centrality of the population, the data point is assigned to the core point set; otherwise, the data point is assigned to the boundary point set. Finally, the internal point set and the boundary point set are obtained.

[0126] The core point secondary evaluation is used to perform a secondary verification of the internal point set, screening out stable core points and eliminating pseudo-core points and noise points, thus providing a reliable starting point for subsequent core cluster expansion. Specifically, the core point secondary evaluation is based on the reverse neighborhood quantity evaluation mechanism. First, for each data point in the internal point set, the reverse k-nearest neighbor set of that data point is calculated, and then the number of data points in the reverse k-nearest neighbor set of that data point is counted. If the data point If k is true, then the data point is determined to be a core point; otherwise, the data point is a pseudo-core point.

[0127] The inverse k-nearest neighbor set specifically refers to the set of internal points. For the target data point, iterate through all its internal points and select all internal points j that include the target data point in their own k-nearest neighbor set. The selected internal points then form the inverse k-nearest neighbor set of the target data point. The formula used is as follows:

[0128] ;

[0129] The core cluster expansion is used to complete the cluster assignment of internal points, construct the core cluster structure of the neurofunctional assessment category, and achieve accurate clustering of different neurofunctional assessment categories. Specifically, firstly, all internal points are marked as unassigned, and each unassigned core point is used as the initial cluster center. Then, the k-nearest neighbor set and the reverse k-nearest neighbor set of the core point are merged as the extended neighborhood. Subsequently, double constraint condition verification is performed on all unassigned internal points in the extended neighborhood. Only points that meet both conditions can be assigned to the cluster of the current core point. Finally, the data point newly added to the cluster is used as the new cluster center. The extended neighborhood and double constraint condition verification are iteratively executed until no new internal data point is added to the current cluster. Finally, all core points are traversed to complete the cluster assignment of all internal points.

[0130] The points within the extended neighborhood are limited to those in the internal point set;

[0131] The dual constraint condition verification specifically includes orientation centrality consistency constraint and reverse neighborhood density constraint;

[0132] The directional centrality consistency constraint specifically means that the adaptive directional centrality measure of the internal point to be assigned and the core point that serves as the cluster center are less than the group average directional centrality measure of the two.

[0133] The reverse neighborhood density constraint requires that the number of data points in the reverse k-nearest neighbor set of the internal points to be assigned be greater than or equal to the minimum neighborhood threshold. ;

[0134] The formula used is as follows:

[0135] ;

[0136] In the formula, This represents the adaptive directional centrality metric of the core point 'a' at the cluster center. This represents the adaptive orientation centrality metric of the interior point b to be assigned. This represents the number of data points in the inverse k-nearest neighbor set of the internal point to be assigned.

[0137] The remaining point allocation is used to complete the cluster assignment of boundary points and unassigned points, and realize the complete clustering of the full neural function assessment. Specifically, for each data point in the boundary point set, the nearest assigned internal point is searched and the cluster label of the internal point is assigned to the current boundary point. Then, for the remaining unassigned data points, the nearest assigned data point is searched and the cluster label of the point is assigned to the current unassigned data point.

[0138] The clustering results are output by merging all data points into the corresponding cluster set after all data points have been assigned, forming multiple independent clusters. Each cluster represents a category of the neurological function assessment results after a patient's spinal cord injury, thus obtaining the clustering output results.

[0139] The pre-injury baseline homogeneity screening is used to screen reference patients' post-injury assessment data that are homogeneous with the target patient's pre-injury neurological function baseline results. This ensures the baseline consistency of the model's input data and completely eliminates the interference of different pre-injury baseline levels on the neurological function assessment results. Specifically, based on the target patient's pre-injury neurological function baseline results as the screening criterion, all reference patients are screened to identify those whose pre-injury neurological function assessment results are the same as the target patient's. The post-injury assessment data of the selected reference patients is then extracted. Finally, the selected reference patient post-injury assessment data and the target patient post-injury assessment data are subjected to feature extraction according to the core feature subset of post-injury neurological function assessment to obtain feature data, which is used as the input data for the post-injury neurological function assessment model.

[0140] Real-time assessment of neurological function after spinal cord injury involves inputting the data from the neurological function assessment model into the model, generating real-time clustering output results, and statistically analyzing the cluster labels in each cluster based on the real-time clustering output results. The cluster label with the highest frequency of occurrence is selected as the neurological function assessment result label for that cluster, thus obtaining the neurological function assessment result of the target patient after spinal cord injury.

[0141] Comprehensive assessment of neurological function after spinal cord injury involves a comprehensive evaluation of the patient's neurological function after spinal cord injury, based on the baseline neurological function results before the injury and the neurological function assessment results after the injury. This aims to achieve a comprehensive, precise, and intelligent assessment of the patient's neurological function after spinal cord injury.

[0142] By performing the above operations, this solution addresses the technical problems of existing clustering algorithms used in post-injury neurological function assessment models, which rely on manual global parameter tuning and have poor compatibility with high-dimensional multimodal data, leading to inaccurate clustering results and insufficient accuracy of model output. This innovative approach improves the clustering algorithm by integrating a high-dimensional vector angle variance quantification method with a core point secondary evaluation strategy based on an inverse neighborhood quantity evaluation mechanism. This completely eliminates subjective errors from manual parameter tuning, accurately adapts to data, is compatible with high-dimensional multimodal data, and significantly improves clustering efficiency and accuracy. This enhances the accuracy of the output results of the post-spinal cord injury neurological function assessment model, providing a stable and repeatable homogeneous population-based cross-sectional benchmark for a dual-benchmark comprehensive assessment system of spinal cord injury neurological function. Ultimately, this achieves a comprehensive, precise, and intelligent assessment of neurological function after spinal cord injury in patients.

[0143] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0144] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0145] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A comprehensive assessment system for neurological function after spinal cord injury based on big data, characterized in that: It includes a data acquisition module, a multimodal feature depth selection module for spinal cord injury, a pre-injury neurological function baseline construction module, and a dual-benchmark spinal cord injury neurological function comprehensive assessment module; The data acquisition module obtains optimized neural function assessment data through data collection and data preprocessing operations. The spinal cord injury multimodal feature depth selection module calculates the ternary mutual information between features and superimposes the ternary mutual information between features on the basis of the initial weights to calculate the final feature weights. Based on the final feature weights, it performs core feature subset selection, thereby completing the construction of the collaborative correction feature selection algorithm. Then, the reference patient pre-injury assessment data and reference patient post-injury assessment data are respectively input into the algorithm to obtain the core feature subsets of pre-injury neurological function assessment and post-injury neurological function assessment. The pre-injury neurological function baseline construction module specifically involves constructing a pre-injury neurological function assessment model, training the model, and finally inputting the target pre-injury neurological function assessment core feature data into the trained model to obtain the target patient's pre-injury neurological function baseline results. The dual-benchmark spinal cord injury neurological function comprehensive assessment module improves the clustering algorithm by integrating a high-dimensional vector angle variance quantification method with a core point secondary assessment strategy based on the inverse neighborhood quantity assessment mechanism, and uses this as a post-injury neurological function assessment model. Then, based on the target patient's pre-injury neurological function baseline results as the screening benchmark, the input data of the post-injury neurological function assessment model is obtained. This data is then input into the post-injury neurological function assessment model to obtain the target patient's post-spinal cord injury neurological function assessment results. Finally, based on the target patient's pre-injury neurological function baseline results and the post-spinal cord injury neurological function assessment results, a comprehensive assessment of post-spinal cord injury neurological function is achieved.

2. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 1, characterized in that: The multimodal feature depth selection module for spinal cord injury specifically includes the following steps: The collaborative correction feature selection algorithm is constructed by including the following steps: To construct a feature screening sample matrix, the reference patient injury assessment data in the neurological function assessment optimization data is divided into a multimodal feature set and a sample standard label set. Based on the multimodal feature set and the sample standard label set, a feature screening sample matrix is ​​constructed. The initial weight calculation for a single feature is specifically based on the ReliefF feature selection algorithm, which calculates the initial weights of the features in the feature selection sample matrix to obtain the initial weights of all features. Calculate the ternary mutual information between features; The final weight calculation of features is specifically based on the ternary mutual information value between features, which corrects the initial weights of the features to obtain the final weights of all features. The core feature subset selection process involves using the average final weight of all features as the adaptive selection threshold, retaining features with a final weight greater than the adaptive selection threshold, removing features with a weight less than or equal to the adaptive selection threshold, and combining all features that meet the selection rules to obtain the core feature subset. The pre-injury neurological function assessment feature screening specifically involves inputting the reference patient's pre-injury assessment data from the optimized neurological function assessment data into a collaborative correction feature screening algorithm to obtain a subset of core features for pre-injury neurological function assessment. The feature selection for post-injury neurological function assessment involves inputting the reference patient post-injury assessment data from the optimized neurological function assessment data into a collaborative correction feature selection algorithm to obtain a subset of core features for post-injury neurological function assessment.

3. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 2, characterized in that: The calculation of ternary mutual information between features specifically involves first discretizing the continuous features in the multimodal feature set using equal-frequency methods, dividing them into M equal-frequency intervals to convert the continuous features into discrete features. Then, for any two different features in the multimodal feature set, the mutual information of a single feature and the joint mutual information of the feature pair are calculated sequentially using the standard label set of the samples. Finally, the ternary mutual information value is calculated for each feature pair, and this process is repeated for all feature pairs in the multimodal feature set to complete the ternary mutual information calculation. The formula used is as follows: ； In the formula, and Let each be represented as any two distinct features in the multimodal feature set. This represents the mutual information between the i-th feature and the label set C. Representation of features and The joint mutual information between the feature pairs formed and the label set C. Representation of features , The ternary mutual information value between the label set C and the label set C. This represents the mutual information between the j-th feature and the label set C.

4. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 1, characterized in that: The pre-injury neurological function baseline construction module specifically includes the following steps: The pre-injury neurological function assessment model is constructed and trained. Specifically, a fully connected neural network is used to build the pre-injury neurological function assessment model. Then, the core feature data of the reference pre-injury neurological function assessment is input into the pre-injury neurological function assessment model for model training, and the trained pre-injury neurological function assessment model is obtained. To obtain the baseline of pre-injury neurological function of the target patient, the core feature data of the pre-injury neurological function assessment are input into the trained pre-injury neurological function assessment model to obtain the pre-injury neurological function assessment results of the target patient, which serve as the baseline results of the pre-injury neurological function of the target patient.

5. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 1, characterized in that: The dual-benchmark spinal cord injury neurological function comprehensive assessment module specifically includes the following steps: Construct a model for assessing neurological function after injury; The pre-injury baseline homogeneity screening is specifically carried out by using the pre-injury neurological function baseline results of the target patient as the screening criterion. All reference patients are screened to select those whose pre-injury neurological function assessment results are the same as those of the target patient. The post-injury assessment data of the selected reference patients are then extracted. The post-injury assessment data of the selected reference patients and the post-injury assessment data of the target patient are then subjected to feature extraction according to the core feature subset of post-injury neurological function assessment to obtain feature data, which is used as the input data of the post-injury neurological function assessment model. Real-time assessment of neurological function after spinal cord injury involves inputting the data from the neurological function assessment model into the model, generating real-time clustering output results, and statistically analyzing the cluster labels in each cluster based on the real-time clustering output results. The cluster label with the highest frequency of occurrence is selected as the neurological function assessment result label for that cluster, thus obtaining the neurological function assessment result of the target patient after spinal cord injury. Comprehensive assessment of neurological function after spinal cord injury involves comprehensively evaluating the neurological function of the patient after spinal cord injury based on the baseline neurological function results before the injury and the neurological function assessment results after the spinal cord injury.

6. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 5, characterized in that: The construction of the post-injury neurological function assessment model specifically includes the following steps: The adaptive orientation centrality metric is calculated by employing a high-dimensional vector angle variance quantization method. For each data point, the Euclidean distance between it and other data points is calculated. All other data points are then sorted in ascending order of their distance values, and the k nearest neighbors with the smallest distances at the top of the sort are selected to form the k-nearest neighbor set for that data point. Then, using that data point as the starting point, high-dimensional space vectors are drawn to any two different data points in the k-nearest neighbor set, and the angle between the two vectors is calculated. This process is repeated for all pairwise distinct point combinations in the k-nearest neighbor set to complete the angle calculation. Finally, the statistical variance of all pairwise angles is calculated to obtain the adaptive orientation centrality metric value for that data point. The population average directional centrality metric is calculated by taking the adaptive directional centrality metric value of each data point as the basis. For each data point, the average of the adaptive directional centrality metric values ​​of all data points in the k nearest neighbor set of that data point is calculated to obtain the population average directional centrality metric of that data point. The data points are initially divided adaptively. Specifically, all data points are traversed, and the adaptive directional centrality metric of each point is compared with the average directional centrality metric of the population. If the adaptive directional centrality metric of the data point is less than the average directional centrality metric of the population, the data point is assigned to the core point set; otherwise, the data point is assigned to the boundary point set. Finally, the internal point set and the boundary point set are obtained. The core point secondary evaluation is specifically based on the reverse neighborhood quantity evaluation mechanism. First, for each data point in the internal point set, the reverse k-nearest neighbor set of that data point is calculated. Then, the number of data points in the reverse k-nearest neighbor set of that data point is counted. If the data point If k is true, then the data point is determined to be a core point; otherwise, the data point is a pseudo-core point. The inverse k-nearest neighbor set specifically refers to the set of internal points. For the target data point, iterate through all internal points and select all internal points j that include the target data point in their own k-nearest neighbor set. The selected internal points form the inverse k-nearest neighbor set of the target data point. Core cluster expansion; The remaining point allocation process involves searching for the nearest allocated internal point for each data point in the boundary point set, assigning the cluster label of that internal point to the current boundary point, and then searching for the nearest allocated data point for the remaining unassigned data points, assigning the cluster label of that point to the current unassigned data point. The clustering results are output by merging all data points into their corresponding cluster sets after all data points have been assigned, forming multiple independent clusters. Each cluster represents a category of a patient's neurological function assessment results after spinal cord injury.

7. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 6, characterized in that: The core cluster expansion process involves first marking all internal points as unassigned, using each unassigned core point as the initial cluster center, then merging the k-nearest neighbor set and the reverse k-nearest neighbor set of the core point as an expanded neighborhood, subsequently performing a double constraint check on all unassigned internal points in the expanded neighborhood, and only points that simultaneously satisfy both conditions can be assigned to the cluster of the current core point, and finally using the newly added data point as the new cluster center, iteratively executing the expanded neighborhood and double constraint check until no new internal data point is added to the current cluster, and finally traversing all core points to complete the cluster partitioning of all internal points; The dual constraint condition verification specifically includes orientation centrality consistency constraint and reverse neighborhood density constraint; The directional centrality consistency constraint specifically means that the adaptive directional centrality measure of the internal point to be assigned and the core point that serves as the cluster center are less than the group average directional centrality measure of the two. The reverse neighborhood density constraint requires that the number of data points in the reverse k-nearest neighbor set of the internal points to be assigned be greater than or equal to the minimum neighborhood threshold. .

8. The comprehensive assessment system for neurological function after spinal cord injury based on big data as described in claim 1, characterized in that: The data acquisition module specifically acquires raw neurological function assessment data by performing data collection operations, and preprocesses the raw neurological function assessment data to obtain optimized neurological function assessment data. The raw neurological function assessment data includes pre-injury assessment data of a reference patient, pre-injury assessment data of a target patient, post-injury assessment data of a reference patient, and post-injury assessment data of a target patient. The data preprocessing specifically involves sequentially performing data cleaning and normalization on the raw neurological function assessment data to obtain optimized neurological function assessment data.

Images