CN-121983295-A - Comprehensive evaluation system for nerve function after spinal cord injury based on big data

CN121983295ACN 121983295 ACN121983295 ACN 121983295ACN-121983295-A

Abstract

The invention discloses a spinal cord injury post-nerve function comprehensive evaluation system based on big data, which comprises a data acquisition module, a spinal cord injury multi-mode feature depth selection module, a pre-injury nerve function baseline construction module and a double-reference spinal cord injury nerve function comprehensive evaluation module. The invention relates to the technical field of medical data processing, in particular to a comprehensive evaluation system of nerve function after spinal cord injury based on big data, which innovatively combines a pre-injury nerve function baseline construction module and a double-benchmark comprehensive evaluation module of nerve function after spinal cord injury to improve the comprehensiveness and accuracy of evaluation results; the clustering algorithm is innovatively improved by fusing a high-dimensional vector included angle variance quantification method with a core point secondary evaluation strategy based on a reverse neighborhood quantity evaluation mechanism, the accuracy of a clustering result is improved, and the accuracy of a neural function evaluation result after spinal cord injury is improved.

Inventors

JU CHENG
HU HUIMIN
LIU RENFENG
ZHAO YUQI
DONG HUI
WANG YUANWEI
XIA ZHIHAO

Assignees

西安市红会医院（西安市骨科研究所）

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (8)

1. The system is characterized by comprising a data acquisition module, a spinal cord injury multi-mode feature depth selection module, a pre-injury nerve function baseline construction module and a double-benchmark spinal cord injury nerve function comprehensive evaluation module; the data acquisition module acquires neural function evaluation optimization data through data acquisition and data preprocessing operation; The spinal cord injury multi-mode feature depth selection module performs feature final weight calculation by calculating ternary mutual information among features, superposing the ternary mutual information among the features on the basis of initial weights, and performing core feature subset screening based on the feature final weights, so that the construction of a collaborative correction feature screening algorithm is completed, and then, evaluation data before injury of a reference patient and evaluation data after injury of the reference patient are respectively input into the algorithm to obtain a pre-injury neural function evaluation core feature subset and a post-injury neural function evaluation core feature subset; the pre-injury nerve function baseline construction module is used for constructing a pre-injury nerve function evaluation model, performing model training, and finally inputting target pre-injury nerve function evaluation core characteristic data into the trained model to obtain a target patient pre-injury nerve function baseline result; The dual-benchmark spinal cord injury nerve function comprehensive evaluation module is used for improving a clustering algorithm by fusing a high-dimensional vector included angle variance quantification method with a core point secondary evaluation strategy based on a reverse neighborhood quantity evaluation mechanism, taking the improved clustering algorithm as a post-injury nerve function evaluation model, taking a target patient pre-injury nerve function baseline result as a screening benchmark, acquiring input data of the post-injury nerve function evaluation model, inputting the data into the post-injury nerve function evaluation model to obtain a target patient spinal cord injury nerve function evaluation result, and finally, realizing spinal cord injury nerve function comprehensive evaluation based on the target patient pre-injury nerve function baseline result and the spinal cord injury nerve function evaluation result.
2. The system for comprehensively evaluating nerve function after spinal cord injury based on big data according to claim 1, wherein the multi-modal feature depth selection module for spinal cord injury specifically comprises the following steps: Constructing a collaborative correction feature screening algorithm, which specifically comprises the following steps: Constructing a feature screening sample matrix, namely dividing reference patient damage evaluation data in nerve function evaluation optimization data into a multi-mode feature set and a sample standard label set, and constructing the feature screening sample matrix based on the multi-mode feature set and the sample standard label set; the single feature initial weight calculation is specifically that based on ReliefF feature selection algorithm, the initial weight calculation is carried out on the features in the feature screening sample matrix to obtain all feature initial weights; calculating ternary mutual information among features; calculating the final weights of the features, specifically correcting the initial weights of the features based on the ternary mutual information values among the features to obtain final weights of all the features; The method comprises the steps of screening a core feature subset, namely taking an average value of final weights of all features as an adaptive screening threshold, reserving the features with the final weights larger than the adaptive screening threshold, eliminating the features with the weights smaller than or equal to the adaptive screening threshold, and combining all the features meeting screening rules to obtain the core feature subset; The pre-injury nerve function evaluation feature screening is carried out, specifically, the pre-injury evaluation data of a reference patient in nerve function evaluation optimization data is input into a collaborative correction feature screening algorithm, and a pre-injury nerve function evaluation core feature subset is obtained; the post-injury nerve function evaluation feature screening is specifically to input post-injury evaluation data of a reference patient in nerve function evaluation optimization data into a collaborative correction feature screening algorithm to obtain a post-injury nerve function evaluation core feature subset.
3. The system for comprehensively evaluating the nerve function after the spinal cord injury based on big data according to claim 2, wherein the system for calculating the ternary mutual information among the characteristics is characterized in that firstly, continuous characteristics in a multi-mode characteristic set are subjected to equal-frequency discretization, divided into M equal-frequency intervals, the continuous characteristics are converted into discrete characteristics, then, any two different characteristics in the multi-mode characteristic set are combined with a sample standard label set to sequentially finish single-characteristic mutual information calculation and characteristic pair combination mutual information calculation, and finally, ternary mutual information values are calculated for each group of characteristic pairs and all characteristic pairs in the multi-mode characteristic set are traversed to finish ternary mutual information calculation, and the formula is as follows: ; In the formula, And Respectively represented as any two mutually different features in the multi-modal feature set, Representing mutual information between the ith feature and the tag set C, Representing characteristics And (3) with The combined mutual information between the composed feature pairs and the tag set C, Representing characteristics 、 Ternary mutual information values with the tag set C, Representing mutual information between the jth feature and the tag set C.
4. The integrated post-injury nerve function evaluation system based on big data according to claim 1, wherein the pre-injury nerve function baseline construction module specifically comprises the following steps: The method comprises the steps of constructing and training a pre-injury nerve function assessment model, specifically, adopting a fully-connected neural network to establish the pre-injury nerve function assessment model, and then inputting reference pre-injury nerve function assessment core characteristic data into the pre-injury nerve function assessment model to perform model training to obtain a trained pre-injury nerve function assessment model; The method comprises the steps of obtaining a target patient pre-injury nerve function baseline, specifically inputting target pre-injury nerve function evaluation core characteristic data into a trained pre-injury nerve function evaluation model, and obtaining a target patient pre-injury nerve function evaluation result as a target patient pre-injury nerve function baseline result.
5. The system for comprehensively evaluating the nerve function after the spinal cord injury based on big data according to claim 1, wherein the double-reference spinal cord injury nerve function comprehensive evaluation module specifically comprises the following steps: constructing a post-injury neural function assessment model; The method comprises the steps of carrying out homogeneous screening on a baseline before injury, specifically, traversing all reference patients, wherein the baseline before injury is used as a screening standard, screening out reference patients, the evaluation result of the nerve function of the reference patients is the same as the baseline result of the nerve function of the target patients before injury, extracting evaluation data after injury of the reference patients corresponding to the reference patients obtained through screening, carrying out feature extraction on the evaluation data after injury of the reference patients obtained through screening and the evaluation data after injury of the target patients, and obtaining feature data after feature extraction according to a core feature subset of the evaluation of the nerve function after injury, and taking the feature data as input data of an evaluation model of the nerve function after injury; The method comprises the steps of performing real-time assessment on the nerve function after the spinal cord injury, namely inputting input data of a nerve function assessment model after the injury into the nerve function assessment model, generating a real-time clustering output result, counting cluster labels in each cluster according to the real-time clustering output result, and selecting the cluster label with the highest occurrence frequency as a nerve function assessment result label after the spinal cord injury of the cluster to obtain a nerve function assessment result after the spinal cord injury of a target patient; The comprehensive evaluation of the nerve function after the spinal cord injury is specifically based on a target patient pre-injury nerve function baseline result and a target patient post-spinal cord injury nerve function evaluation result, so that the comprehensive evaluation of the nerve function after the spinal cord injury of the patient is realized.
6. The system for comprehensively evaluating post-injury nerve function based on big data according to claim 5, wherein the method for constructing the post-injury nerve function evaluation model specifically comprises the following steps: Calculating self-adaptive direction centrality measurement, namely calculating Euclidean distance of each data point and other data points by adopting a high-dimensional vector included angle variance quantification method, sorting all other data points in ascending order according to distance values, taking k data points with the smallest distances in the front of the sorting order to form a k neighbor set of the data points, taking the data points as vector starting points, respectively making high-dimensional space vectors to any two different data points in the k neighbor set, calculating included angles between the two vectors, traversing all two-to-two different point combinations in the k neighbor set to finish calculation of the included angles, and finally calculating statistical variances of all two-to-two included angles to finally obtain the self-adaptive direction centrality measurement value of the data points; Calculating group average direction centrality measurement, specifically calculating the average value of all data point self-adaptive direction centrality measurement values in a k neighbor set of each data point based on the self-adaptive direction centrality measurement value of each data point to obtain group average direction centrality measurement of the data point; The data point self-adaptive preliminary division specifically comprises the steps of traversing all data points, comparing the self-adaptive direction centrality measurement value of each data point with the group average direction centrality measurement value, if the self-adaptive direction centrality measurement value of the data point is smaller than the group average direction centrality measurement value, classifying the data point into a core point set, otherwise classifying the data point into a boundary point set, and finally obtaining an internal point set and a boundary point set; Core point secondary evaluation, specifically, core point secondary evaluation is performed based on a reverse neighborhood number evaluation mechanism, firstly, for each data point in an internal point set, calculating a reverse k neighbor set of the data point, and then counting the number of data points of the reverse k neighbor set of the data point If the data point K, judging the data point as a core point, otherwise, judging the data point as a pseudo core point; The reverse k neighbor set is specifically a set of internal points Traversing all internal points, screening out all internal points j which bring the target data point into a k neighbor set of the internal points, and forming a reverse k neighbor set of the target data point by all screened internal points; expanding a core cluster; the method comprises the steps of distributing residual points, namely searching for each data point in a boundary point set, namely searching for the nearest distributed internal point, assigning a cluster label of the internal point to a current boundary point, searching for the data point of which the label is not distributed, and assigning the cluster label of the point to the current unassigned data point; And outputting a clustering result, namely merging all data points into corresponding clustering after the distribution of the data points is completed, forming a plurality of mutually independent clusters, wherein each cluster represents the category of the neural function evaluation result after the spinal cord injury of one patient, and the obtained clustering output result.
7. The system for comprehensively evaluating the nerve function after the spinal cord injury based on big data according to claim 6, wherein the core cluster is expanded, namely firstly, cluster labels of all inner points are marked as unassigned, each unassigned core point is used as an initial cluster center, then a k neighbor set and a reverse k neighbor set of the core point are combined to be used as an expansion neighborhood, then double constraint condition verification is carried out on all unassigned inner points in the expansion neighborhood, only points meeting two conditions can be classified into the cluster where the current core point is located, finally, the data points newly added into the cluster are used as new cluster centers, the expansion neighborhood and the double constraint condition verification are iteratively executed until no new inner data points are added into the current cluster, and finally all the core points are traversed, so that cluster division of all the inner points is completed; The double constraint condition verification specifically comprises a directional centrality consistency constraint and a reverse neighborhood density constraint; the direction centrality consistency constraint is that the self-adaptive direction centrality measurement of an internal point to be distributed and a core point serving as a cluster center is smaller than the group average direction centrality measurement of the internal point to be distributed and the core point serving as the cluster center; The number of data points of the reverse k neighbor set of the internal points to be distributed is required to be more than or equal to the minimum neighborhood threshold value 。
8. The system for comprehensively evaluating the nerve function after the spinal cord injury based on big data according to claim 1, wherein the data acquisition module is used for acquiring nerve function evaluation original data through data acquisition operation, and performing data preprocessing on the nerve function evaluation original data to acquire nerve function evaluation optimization data, wherein the nerve function evaluation original data comprises reference patient pre-injury evaluation data, target patient pre-injury evaluation data, reference patient post-injury evaluation data and target patient post-injury evaluation data, and the data preprocessing is used for sequentially performing data cleaning and normalization processing on the nerve function evaluation original data to acquire the nerve function evaluation optimization data.

Description

Comprehensive evaluation system for nerve function after spinal cord injury based on big data Technical Field The invention relates to the technical field of medical data processing, in particular to a comprehensive evaluation system for nerve function after spinal cord injury based on big data. Background The comprehensive evaluation system for the nerve function after the spinal cord injury is an intelligent management system for the medical health information of the whole period of rehabilitation of a spinal cord injury patient, and by collecting the multi-source data of the whole period of the spinal cord injury patient and relying on the medical big data processing and artificial intelligent information analysis technology, the comprehensive evaluation and dynamic monitoring of informatization are carried out on the nerve function injury state and the recovery progress of the spinal cord injury patient, so that the dynamic tracking of the nerve function recovery, the accurate evaluation of the rehabilitation curative effect and the clinical decision support can be provided for a medical management main body, and the whole period intelligent information processing, the accurate evaluation and the standardized management of the nerve function rehabilitation after the spinal cord injury are realized. However, the traditional comprehensive evaluation system for the nerve function after the spinal cord injury only adopts a single group grading standard after the injury, so that the technical problems of single dimension and inaccurate evaluation result of the comprehensive evaluation of the nerve function after the spinal cord injury are caused; the conventional feature selection algorithm can only perform single feature weight evaluation and ignore interaction effects among multi-modal features, so that the technical problems of feature importance quantification deviation, incomplete redundant feature elimination and underestimation of high-collaborative value features are caused, and the existing clustering algorithm suitable for the post-injury neural function evaluation model has the technical problems of inaccurate clustering results and insufficient accuracy of model output results due to dependence on artificial global parameter tuning and poor high-dimensional multi-modal data calculation compatibility. Disclosure of Invention Aiming at the technical problems that the post-injury nerve function comprehensive evaluation dimension is single and the evaluation result is inaccurate due to the fact that only a single post-injury group grading standard is adopted in the traditional post-injury nerve function comprehensive evaluation system, the technical scheme creatively combines a pre-injury nerve function baseline construction module and a double-standard spinal cord injury nerve function comprehensive evaluation module, takes an exclusive baseline before injury as a first standard and takes a baseline homogeneous group clustering result as a second standard, simultaneously considers the longitudinal function change before and after injury and the homogeneous group transverse level positioning of an individual, effectively overcomes the inherent defect of the traditional single evaluation standard, improves the comprehensiveness and accuracy of the post-injury nerve function grading evaluation, and realizes the integrated comprehensive evaluation of the longitudinal individual change quantification and the transverse group level positioning of the spinal cord injury nerve function; aiming at the technical problems that the traditional feature selection algorithm can only evaluate single feature weights and ignore interaction effects among multi-mode features, thereby causing feature importance quantification deviation, incomplete redundant feature elimination and underestimation of high-collaborative value features, and further causing insufficient precision of an evaluation model, the invention innovatively provides a collaborative correction feature screening algorithm integrating ternary mutual information, performs feature final weight calculation by superposing the ternary mutual information among features on the basis of initial weights, constructs a self-adaptive screening threshold based on the feature final weights to screen a core feature subset, effectively improves the accuracy and rationality of feature importance sorting, the method has the advantages that the mining capability of correlation and synergistic effect among the multi-modal features is enhanced, the filtering efficiency of the redundant features and the noise features is improved, the accuracy, the self-adaption and the Gao Panbie force optimization of the multi-modal features in a spinal cord injury scene are realized, the simplified and efficient feature input is provided for a follow-up neural function evaluation model, the evaluation precision and the stability of the model are improved, the technical problems that the cl