Intelligent analysis system and method for rehabilitation treatment of lumbar intervertebral disc protrusion

Abstract

The invention discloses an intelligent analysis system and method for rehabilitation treatment of lumbar intervertebral disc protrusion, and relates to the technical field of medical informationization, and the system comprises a multi-modal data collection module which is used for obtaining multi-modal medical data of a patient; the feature adaptive optimization module is used for screening a rehabilitation prediction feature set and dynamically optimizing a feature weight; the rehabilitation prediction analysis module is used for generating a rehabilitation process prediction result; the training scheme generation module is used for generating a rehabilitation training scheme; the feedback data acquisition module is used for monitoring feedback data in the training process; the rehabilitation evaluation and adjustment module is used for evaluating a rehabilitation process state and dynamically adjusting a rehabilitation training scheme; and the incremental learning updating module is used for updating the feature correlation prediction model. Through mutual cooperation and mutual promotion of the functional modules, the precision, intelligence and standardization level of lumbar intervertebral disc protrusion rehabilitation treatment is improved, and powerful technical support is provided for improvement of the rehabilitation treatment effect.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to an intelligent analysis system and method for rehabilitation treatment of lumbar disc herniation. Background Art

[0002] Currently, a variety of intelligent assistance systems have been developed for the rehabilitation treatment of lumbar disc herniation. These primarily use wearable devices to collect patient motion data and combine it with traditional machine learning algorithms to assess rehabilitation status and provide training guidance. Some systems also incorporate medical image recognition technology to assist in assessing patients' recovery progress. These systems have initially achieved digital recording and basic analysis of the rehabilitation training process, providing valuable insights into the intelligentization of rehabilitation treatment.

[0003] However, existing systems still have the following technical limitations: Feature extraction strategies are relatively fixed, making them difficult to adapt to the characteristics of different rehabilitation stages; rehabilitation prediction models use a unified model structure that fails to fully account for individual patient differences; and training program adjustments are primarily based on single-dimensional evaluation metrics, lacking comprehensive effect evaluation and systematic optimization strategies. These technical limitations have impacted the practical application of intelligent rehabilitation assistance systems. Summary of the Invention

[0004] The present invention provides an intelligent analysis system and method for the rehabilitation treatment of lumbar disc herniation, which is used to solve the technical problems of insufficient feature optimization, poor prediction accuracy and unsystematic program adjustment in existing rehabilitation intelligent auxiliary systems.

[0005] In view of this, the first aspect of the present invention provides an intelligent analysis system for rehabilitation treatment of lumbar disc herniation, comprising: Multimodal data acquisition module, used to obtain multimodal medical data of patients with lumbar disc herniation; The feature adaptive optimization module is used to preprocess multimodal medical data, apply a feature importance adaptive ranking algorithm to screen the recovery prediction feature set, and dynamically optimize feature weights based on the patient's recovery stage; The rehabilitation prediction analysis module is used to analyze the rehabilitation prediction feature set based on the feature association prediction model and generate the patient's rehabilitation process prediction results; A training program generation module is used to generate a rehabilitation training program based on the rehabilitation process prediction results through a multi-objective optimization algorithm; Feedback data collection module, used to monitor the feedback data of patients during the implementation of rehabilitation training programs; The rehabilitation assessment and adjustment module is used to evaluate the status of the rehabilitation process based on the multi-dimensional rehabilitation evaluation indicator system and feedback data, and dynamically adjust the rehabilitation training plan according to the assessment results; The incremental learning update module is used to update the feature association prediction model through incremental learning based on the difference between the execution results of the adjusted rehabilitation training program and the predicted results of the rehabilitation process.

[0006] Optionally, the multimodal medical data includes clinical assessment data, imaging data, rehabilitation training data, and daily activity monitoring data.

[0007] Optionally, the processing flow of the feature adaptive optimization module includes: Extracting candidate feature sets from preprocessed multimodal medical data; A combined feature evaluation method is used to calculate the initial importance score of each candidate feature and generate a feature priority list; Construct feature correlation matrix, calculate mutual information and correlation coefficient between features, and identify redundant features and complementary features; Combined with the results of feature correlation analysis, the feature priority list is adjusted for correlation penalties to establish a feature benchmark ranking; Establish a stage feature mapping relationship based on the patient's recovery stage and dynamically adjust the feature weight coefficient; Based on the adjusted feature weights and feature benchmark rankings, the optimal feature subset is selected from the candidate feature set to generate the rehabilitation prediction feature set.

[0008] Optionally, the processing flow of the rehabilitation prediction analysis module includes: Design the infrastructure for feature association prediction models; A multidimensional clinical feature classification strategy was used to divide patients into multiple disc herniation subgroups; Based on a historical patient rehabilitation database, the feature association prediction model was pre-trained through transfer learning, and the Bayesian optimization method was used to automatically select the optimal model parameters for each subgroup of patients; Design a multi-objective joint learning framework to simultaneously train three prediction tasks: functional recovery, symptom relief, and relapse risk, sharing the underlying feature representation; Evaluate model performance and optimize the model's predictive calibration using Bayesian posterior probability calibration techniques combined with clinical decision thresholds; Output the patient's recovery process prediction results.

[0009] Optionally, the processing flow of the training plan generation module includes: Build a rehabilitation training action library and establish a dual attribute mapping including rehabilitation goals and risk levels for each training action; Construct a multi-objective optimization function and formulate phased rehabilitation goals and evaluation criteria; The training parameter limits are determined based on the staged rehabilitation goals and rehabilitation process prediction results, and a risk-constrained hierarchical search strategy is used to select candidate action combinations from the rehabilitation training action library. The analytic hierarchy process was applied to prioritize candidate action combinations and generate the optimal action sequence, and Bayesian optimization was used to determine the rehabilitation training prescription parameters for each training action. Integrate the optimal movement sequence, training prescription parameters and phased rehabilitation goals to generate a rehabilitation training plan.

[0010] Optionally, the processing flow of the rehabilitation assessment and adjustment module includes: Construct a multidimensional rehabilitation evaluation indicator system; Apply multi-perspective temporal correlation analysis methods to analyze the mutual influence and temporal evolution patterns among evaluation indicators and establish an indicator correlation map; A hierarchical Bayesian model was used to integrate the indicator association map with the patient's historical rehabilitation data to determine the weight coefficients of each evaluation indicator at different rehabilitation stages; Mapping patient feedback data to a multidimensional rehabilitation evaluation index system to generate standardized scores for the corresponding evaluation indicators; Compare and analyze the standardized scores with the phased goals of the training program and calculate the target achievement of each evaluation indicator; Apply weight coefficients to the target achievement of each evaluation indicator for weighted calculation, analyze the impact links based on the indicator association map, and identify priority adjustment indicators; The target achievement of each evaluation indicator is weighted by applying the weight coefficient, and the impact link is analyzed based on the indicator association map to identify the priority adjustment indicators.

[0011] Optionally, the processing flow of the incremental learning update module includes: Calculate the deviation index between the patient's actual rehabilitation execution results and the predicted rehabilitation process, and determine the main characteristic dimensions of the prediction deviation; Determine the model features and parameters that need to be optimized based on the deviation index analysis; Extract historical data related to the main feature dimensions from the patient rehabilitation database to construct an incremental learning dataset; Update the parameters of the feature association prediction model and adjust the model parameters related to the main feature dimensions.

[0012] A second aspect of the present invention provides an intelligent analysis method for rehabilitation treatment of lumbar disc herniation, comprising: Acquire multimodal medical data of patients with lumbar disc herniation; Preprocess multimodal medical data, apply feature importance adaptive ranking algorithm to screen rehabilitation prediction feature set, and dynamically optimize feature weights based on the patient's recovery stage; Construct a feature association prediction model, input the rehabilitation prediction feature set into the feature association prediction model, and obtain the patient's rehabilitation process prediction result; Based on the rehabilitation process prediction results, a multi-objective optimization algorithm is used to select action combinations and optimize training parameters to generate a rehabilitation training plan. Monitor patient feedback during digital rehabilitation programs; Construct a multi-dimensional rehabilitation evaluation indicator system to assess the status of rehabilitation progress based on feedback data, and dynamically adjust rehabilitation training programs based on the evaluation results; Based on the difference between the execution results of the adjusted rehabilitation training program and the predicted results of the rehabilitation process, the feature association prediction model is updated through incremental learning.

[0013] The beneficial effects of the present invention are as follows: through feature adaptive optimization and patient classification prediction, the present invention can realize accurate customization of rehabilitation plans according to individual characteristics of patients and characteristics of rehabilitation stages, improve the matching degree between training plans and patient status, and provide strong support for personalized rehabilitation treatment; based on a multidimensional rehabilitation evaluation index system and feedback data, a real-time monitoring and dynamic adjustment mechanism of the rehabilitation process is established, which can timely identify deviations in rehabilitation effects, dynamically optimize training plans, and ensure the continued effectiveness of rehabilitation treatment; through comprehensive analysis of multimodal medical data and continuous optimization of models, the subjectivity in the process of formulating rehabilitation plans is reduced, and reliable decision-making support is provided for clinical rehabilitation treatment; combined with the risk level assessment and dynamic monitoring mechanism of training movements, the safety control of training plans is achieved, potential risks can be identified and avoided in a timely manner, and the safety of the rehabilitation training process is ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0015] Figure 1 This is a module connection diagram of an intelligent analysis system for rehabilitation treatment of lumbar disc herniation.

[0016] Figure 2 This is a processing flow chart of the feature adaptive optimization module of an intelligent analysis system for rehabilitation treatment of lumbar disc herniation.

[0017] Figure 3 A processing flow chart for the training program generation module of an intelligent analysis system for the rehabilitation treatment of lumbar disc herniation. DETAILED DESCRIPTION

[0018] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] Reference Figures 1 to 3 , is an embodiment of the present invention, which provides an intelligent analysis system for rehabilitation treatment of lumbar disc herniation, and the module connection diagram is as follows Figure 1 As shown, the system includes: The multimodal data acquisition module is used to obtain multimodal medical data of patients with lumbar disc herniation. The multimodal medical data includes clinical assessment data, imaging data, rehabilitation training data and daily activity monitoring data.

[0020] The feature adaptive optimization module is used to preprocess multimodal medical data, apply the feature importance adaptive ranking algorithm to screen the rehabilitation prediction feature set, and dynamically optimize the feature weights based on the patient's rehabilitation stage.

[0021] The rehabilitation prediction analysis module is used to analyze the rehabilitation prediction feature set based on the feature association prediction model and generate the patient's rehabilitation process prediction results.

[0022] The training program generation module is used to generate a rehabilitation training program based on the rehabilitation process prediction results through a multi-objective optimization algorithm.

[0023] The feedback data acquisition module is used to monitor the feedback data of patients during the execution of rehabilitation training programs. It can be achieved through wearable sensor technology, visual tracking system, bioelectric signal monitoring and other technical means.

[0024] The rehabilitation assessment and adjustment module is used to evaluate the status of the rehabilitation process based on the multidimensional rehabilitation evaluation indicator system and feedback data, and dynamically adjust the rehabilitation training plan according to the assessment results.

[0025] The incremental learning update module is used to update the feature association prediction model through incremental learning based on the difference between the execution results of the adjusted rehabilitation training program and the predicted results of the rehabilitation process.

[0026] Optimally, through the combined collection of clinical assessment data, imaging data, rehabilitation training data, and daily activity monitoring data, a comprehensive characterization of the disease progression in patients with lumbar disc herniation is achieved. Clinical assessment data is used to characterize the patient's subjective and objective symptoms; imaging data is used to characterize changes in the anatomical structure of the lesion; rehabilitation training data is used to characterize the dynamic changes in the treatment process; and daily activity monitoring is used to characterize changes in functional status outside of medical institutions. The combined collection of multimodal data not only ensures the richness and completeness of data features, but also improves data reliability through complementarity between different dimensions.

[0027] The processing flow chart of the feature adaptive optimization module is as follows: Figure 2 As shown, the process includes preprocessing multimodal medical data. For example, clinical assessment data is standardized and outlier correction is performed, imaging data is registered and ROI extraction is performed, rehabilitation training data is subjected to time series feature extraction and quality control, and daily activity monitoring data is subjected to signal filtering and verification. Candidate feature sets are extracted from the preprocessed multimodal medical data, using specialized feature extraction methods for different modalities. The resulting candidate feature sets include, but are not limited to, clinical symptom features, functional status features, imaging features, biomechanical features, training response features, and patient baseline features.

[0028] Furthermore, a combined feature evaluation method is used to calculate the initial importance score of each candidate feature and generate a feature priority ranking list. The specific steps are as follows: perform feature type analysis on the extracted candidate feature set to identify features suitable for entropy-based evaluation and features suitable for gradient-based evaluation; perform entropy-based information analysis and gradient-based sensitivity calculation to obtain the importance index of each candidate feature under the two evaluation systems respectively; based on the feature type and data modal characteristics, determine the information-sensitivity fusion weight coefficient for each candidate feature, and calculate the comprehensive importance score through adaptive weighting; sort the candidate features according to the comprehensive importance score to generate a feature priority ranking list.

[0029] Furthermore, a feature correlation matrix is ​​constructed, and mutual information and correlation coefficients between features are calculated to identify redundant and complementary features. The specific steps are as follows: Pearson correlation coefficients (to capture linear relationships) and normalized mutual information values ​​(to capture nonlinear dependencies) are calculated between all feature pairs in the feature priority list. The calculated correlation indicators are organized into a feature correlation matrix, where each element represents the degree of correlation between the corresponding feature pair. A sparse matrix representation is used to optimize storage for large feature sets. Within the feature correlation matrix, a quartile method based on feature distribution is applied to identify highly correlated feature pairs as potentially redundant. Conditional mutual information is used to analyze the joint contribution of feature combinations to the prediction results. When the individual feature importance is low but the conditional mutual information value is significant, the corresponding feature pair is marked as complementary. High correlation and significant conditional mutual information values ​​are both determined by thresholds.

[0030] Optionally, for categorical feature pairs, the Cramer's coefficient is additionally calculated; for mixed feature pairs (continuous and categorical), the point-biserial correlation coefficient is calculated.

[0031] Preferably, combined with the result of feature correlation analysis, the feature priority list is adjusted for correlation penalty, and a feature benchmark ranking is established. The specific steps are as follows: based on the constructed feature correlation matrix, the redundancy intensity index of each feature pair is calculated, and the index is obtained by weighted fusion of the Pearson correlation coefficient and the normalized mutual information value; according to the redundancy intensity index, a nonlinear penalty function is set, and a hierarchical penalty mechanism is implemented, in which a strong penalty is imposed on high-redundancy features, a moderate penalty is imposed on medium-redundancy features, and a weak penalty is imposed on low-redundancy features, and a redundancy penalty coefficient is generated for each feature, wherein the feature redundancy level is determined by dividing the statistical distribution characteristics of the redundancy intensity index.

[0032] Furthermore, for the identified complementary feature pairs, the complementary reward factor is calculated based on the difference between the joint mutual information of the feature pairs and the sum of the individual mutual information to quantify the degree of complementary gain between the features; the priority score of each feature is recalculated through a weighted adjustment formula based on the initial importance score, redundancy penalty coefficient and complementary reward factor, so as to achieve automatic downgrading of high-correlation feature groups and retention of complementary features; the candidate features in the feature priority ranking list are reordered based on the adjusted priority score to generate a feature benchmark ranking.

[0033] Specifically, a phase-feature mapping relationship is established based on the patient's rehabilitation stage (acute, recovery, and stable phases), and the feature weight coefficients are dynamically adjusted to adapt the feature selection process to the disease progression. In the treatment of lumbar disc herniation, the importance of different features changes dynamically at different rehabilitation stages. For example, the acute phase places greater emphasis on symptom control, the recovery phase places greater emphasis on functional recovery, and the stable phase places greater emphasis on functional maintenance. Dynamically adjusting feature weights based on rehabilitation phase not only reflects the actual needs of clinical medicine but also improves the sensitivity of the rehabilitation prediction model to changes in features at different stages.

[0034] Furthermore, based on the adjusted feature weights and feature benchmark rankings, a recursive feature elimination algorithm is applied to select the optimal feature subset from the candidate feature set, where the recursive feature elimination process is guided by the comprehensive importance score of the feature weights and feature benchmark rankings; weight optimization is performed on the optimal feature subset to generate a rehabilitation prediction feature set.

[0035] Optimally, the feature adaptive optimization module constructs a dynamic, adaptive feature selection and optimization framework. This effectively addresses the issues of feature masking and static selection limitations through the synergistic effects of combined feature evaluation, relevance penalty, and rehabilitation stage mapping. This framework automatically adjusts feature weights based on the patient's rehabilitation stage, ensuring information completeness while eliminating redundancy. This significantly improves the prediction accuracy and clinical utility of the model across different rehabilitation stages.

[0036] The processing flow of the rehabilitation prediction analysis module includes: designing the basic architecture of the feature association prediction model, including the feature extraction layer, multimodal fusion layer, attention interaction layer, time series analysis layer, and prediction output layer.

[0037] First, the feature extraction layer realizes feature extraction of multimodal medical data, converts the multimodal medical data into initial feature representation through a modality-specific encoder, and performs normalization to eliminate scale differences between different data sources; the multimodal fusion layer processes features from different sources through an adaptive weight allocation mechanism to achieve alignment and integration of multimodal features, combines the residual connection structure and layer normalization technology to enhance model stability, and introduces a feature weighting strategy based on clinical knowledge constraints to strengthen the weights of medical-related features.

[0038] Secondly, the attention interaction layer captures the mutual influence between features based on the multi-head self-attention mechanism and the cross-feature attention mechanism; the sequence encoder structure is combined with the temporal information coding technology to process the time dependency in the rehabilitation process; the multi-scale temporal feature extraction framework is used to simultaneously capture short-term changes and long-term trends, and the weight distribution of temporal features is dynamically adjusted according to the patient's rehabilitation stage.

[0039] Finally, the multi-task prediction layer designs the prediction head structure based on the probabilistic prediction framework, adopts the uncertainty quantification learning method, and simultaneously outputs dynamic prediction results of three or more key dimensions: functional recovery trend, symptom relief trajectory, and relapse risk prediction.

[0040] Preferably, the present invention improves the feature extraction quality and association mining capabilities of medical data of different modalities through a feature normalization module and an attention calculation mechanism, while achieving collaborative optimization of prediction accuracy and clinical interpretability based on an uncertainty quantification learning method and a feature weighting strategy based on clinical knowledge constraints.

[0041] In one embodiment, a multi-dimensional clinical feature classification strategy is used to stratify patients into multiple disc herniation subgroups. For example, the multiple disc herniation subgroups can be divided based on anatomical features, including but not limited to a central disc herniation subgroup, a paracentral disc herniation subgroup, a posterolateral disc herniation subgroup, and a distal disc herniation subgroup.

[0042] Furthermore, the multidimensional clinical feature classification strategy can also be used to refine patient classification by combining the most clinically influential dimensions, such as acute, subacute, and chronic subgroups based on disease course, and nerve root compression and non-nerve root compression subgroups based on clinical manifestations. Based on these classification results, a patient subgroup annotation system was constructed to serve as clinical knowledge to guide feature fusion and predictive modeling.

[0043] In another embodiment, based on the historical patient rehabilitation database, the feature association prediction model is pre-trained by transfer learning, and the Bayesian optimization method is used to automatically select the optimal model parameters for each subgroup of patients. The specific steps are as follows: a multi-level pre-training dataset system is constructed, and the historical patient rehabilitation database is divided into a general spinal disease dataset, a lumbar disease dataset, and a lumbar disc herniation-specific dataset according to disease relevance; the feature encoder is trained on the general spinal disease dataset, and the model parameters are pre-trained using a self-supervised learning method to construct a basic feature representation in the field of spinal diseases; selective parameter updates are performed on the lumbar disease dataset, keeping the parameters of specific network layers unchanged, and adapting the model to capture lumbar-specific biomechanical and clinical characteristics; target task adaptive training is performed on the lumbar disc herniation-specific dataset to adapt the model parameters to the clinical manifestations and rehabilitation characteristics of the target disease; the search space of the Bayesian optimization process is defined; Bayesian optimization is performed separately for each disc herniation subgroup, and the optimal hyperparameter combination is efficiently searched through the sequential model optimization algorithm to automatically select the most suitable model configuration for each subgroup.

[0044] Optimally, transfer learning pre-training based on a historical patient rehabilitation database improved the model's predictive capabilities for subgroups with limited data, while Bayesian optimization automatically explored optimal parameter combinations, overcoming the high computational cost of traditional grid search methods. This combination of transfer learning and automatic parameter optimization not only accelerated the convergence of models for each disc herniation subgroup but also effectively prevented overfitting while maintaining predictive accuracy, making it particularly suitable for complex scenarios with unbalanced clinical data.

[0045] In another embodiment, a multi-objective joint learning framework is designed to simultaneously train three prediction tasks of functional recovery, symptom relief and recurrence risk, share the underlying feature representation, and apply a loss function weighting strategy that is dynamically adjusted based on the rehabilitation stage. The specific steps are as follows: adaptive optimization objectives are defined for the three tasks of functional recovery trend, symptom relief trajectory and recurrence risk prediction, where functional recovery adopts mean square error loss, symptom relief adopts weighted cross entropy loss, and recurrence risk adopts time series survival analysis loss; a rehabilitation process status evaluation module is constructed to integrate time series features and dynamic symptom change features, and automatically classify patients into different rehabilitation states through supervised classification methods, including but not limited to acute stage, recovery stage or stable stage; a dynamic task priority allocation mechanism based on the rehabilitation stage is designed to adaptively adjust the importance weight of each prediction task for different rehabilitation states.

[0046] Furthermore, a task-specific feature contribution assessment module was implemented, identifying the most relevant feature combinations for each prediction task through a trainable feature weight assignment mechanism. A multi-stage optimization training strategy, including a task-specific initialization phase, an inter-task knowledge integration phase, and a global parameter optimization phase, was employed to improve the synergy among the model components through progressive training. The combination of transfer learning and automatic parameter optimization not only accelerated the convergence of the models for each disc herniation subgroup, but also effectively prevented overfitting while maintaining predictive accuracy, making it particularly suitable for complex scenarios with unbalanced clinical data.

[0047] Preferably, a dynamic task priority allocation mechanism based on the rehabilitation stage can provide more accurate personalized rehabilitation trajectory predictions compared to traditional static weight models, thereby improving the relevance and reliability of the model in clinical applications. Through task-specific feature contribution evaluation modules and soft parameter sharing mechanisms, the inherent correlations between tasks can be fully utilized, reducing the amount of required training data while improving prediction accuracy.

[0048] Optionally, stratified K-fold cross-validation is used to evaluate model performance, and the model's predictive calibration is optimized using Bayesian posterior probability calibration techniques combined with a clinical decision threshold. It should be noted that Bayesian posterior probability calibration involves adjusting the original predicted probabilities to match the actual observed frequencies using isothermal regression or Bayesian binomial calibration methods, while the clinical decision threshold is determined based on a risk-benefit balance analysis.

[0049] Furthermore, the patient's rehabilitation process prediction results are output, including functional recovery trends, symptom relief trajectories and relapse risk predictions, as well as corresponding confidence intervals.

[0050] Preferably, this step solves the clinical heterogeneity challenges faced by patients with lumbar disc herniation in rehabilitation prediction through the organic combination of the hierarchical design of the feature association prediction model and the clinical feature classification strategy; at the same time, the coordinated application of the multi-objective joint learning framework and the dynamic adjustment mechanism of the rehabilitation stage realizes the unified prediction and dynamic optimization of the three dimensions of functional recovery, symptom relief and recurrence risk.

[0051] The processing flow chart of the training plan generation module is as follows: Figure 3 As shown, it includes: constructing a rehabilitation training movement library and establishing a dual attribute mapping for each training movement, including rehabilitation goals and risk levels. It should be noted that the rehabilitation training movement library includes the functional classification, parameterized description and safety threshold annotation of each training movement. The safety threshold annotation calculates the intervertebral disc pressure value of each movement through a biomechanical model and determines the safety limit range in combination with clinical research data; the dual attribute mapping is constructed by statistically analyzing patient rehabilitation data and an expert scoring system, quantifying the contribution of training movements to rehabilitation goals such as pain relief and functional recovery into a score of 0-10, and dividing the risk levels into three levels based on the risk assessment results of the movements for different types of protrusions.

[0052] It should be noted that the dual attribute mapping is designed to balance the effects and risks associated with lumbar disc herniation rehabilitation. For example, Williams flexion exercises are effective in relieving symptoms but may increase intradiscal pressure. McKenzie stretching is effective for some patients but may worsen symptoms for central herniation. Safety threshold labeling is considered because patients with lumbar disc herniation are extremely sensitive to improper movements, and different movements exert significant variations in pressure on the disc.

[0053] Specifically, based on clinical guidelines and action attribute mapping, a multi-objective optimization function is constructed, including: constructing a function improvement objective function based on a standardized functional assessment scale, constructing a symptom control objective function based on a symptom scoring system, and constructing a safety constraint indicator function based on adverse reaction grading standards; applying the hierarchical analysis method to determine the weight coefficient of each objective function to form a multi-objective optimization function of weighted linear combination; establishing a set of constraint conditions according to clinical safety restrictions to ensure that the program operates within the safety boundary.

[0054] Furthermore, based on the multi-objective optimization function and combined with the clinical rehabilitation staging standards, phased rehabilitation goals and evaluation standards are formulated, including training execution goals, functional improvement goals, symptom control goals and safety monitoring goals. The evaluation standards adopt a quantitative indicator system, through baseline relative values, percentage improvement, standardized scale scores and safety threshold setting, to achieve a multi-dimensional and accurate evaluation of the rehabilitation process.

[0055] Furthermore, the training parameter restriction range is determined according to the stage-by-stage rehabilitation goals and the rehabilitation progress prediction results, and a risk-constrained hierarchical search strategy is adopted to select candidate action combinations from the rehabilitation training action library. The specific steps are as follows: based on the stage-by-stage rehabilitation goals and the rehabilitation progress prediction results, the training parameter restriction range is determined; the rehabilitation training action library is stratified according to the risk level to establish low-, medium-, and high-risk action sets; starting from the low-risk action set, candidate actions that meet the stage-by-stage rehabilitation goals are selected; if the stage-by-stage rehabilitation goals are not met, the search range is gradually expanded to the high-risk action set; within each risk action set, combination screening is performed based on dual attribute mapping to ensure that the candidate action combination covers the stage-by-stage rehabilitation goals.

[0056] Furthermore, the hierarchical analysis method was applied to prioritize the candidate action combinations and generate the optimal action sequence. Bayesian optimization was used to determine the rehabilitation training prescription parameters for each training action in the optimal action sequence. The specific steps are as follows: a hierarchical analysis decision matrix was constructed based on the multi-objective optimization function to determine the evaluation criteria for the action combination priority, including dimensions such as training effect, safety, functional coverage, and action synergy; the weights of each evaluation criterion were calculated by the paired comparison method to generate a comprehensive score for the action combination; the candidate action combinations were prioritized according to the comprehensive score to generate the optimal action sequence; the Bayesian optimization method was used to iteratively optimize within the training parameter limit to determine the training prescription parameter configuration for each training action in the optimal action sequence, including action intensity, training frequency, duration, and difficulty progression strategy.

[0057] Furthermore, the optimal action sequence, training prescription parameters and phased rehabilitation goals are integrated to generate a digital rehabilitation training plan, including a training plan schedule, execution standards and safety warning indicators.

[0058] Optimally, by establishing a dual attribute mapping of training movements and combining risk-constrained hierarchical search and multi-level optimization strategies, a dynamic balance between effectiveness and safety in lumbar disc herniation rehabilitation training is achieved. This technical solution maximizes rehabilitation effectiveness while ensuring safety. Through risk stratification, safer movements are prioritized, and riskier movement combinations are considered only when necessary. Simultaneously, personalized optimization is performed based on the patient's specific circumstances, avoiding potential safety risks associated with standardized protocols. This effectively addresses the effectiveness-safety trade-off in lumbar disc herniation rehabilitation training.

[0059] The processing flow of the rehabilitation assessment and adjustment module includes: constructing a multidimensional rehabilitation evaluation indicator system corresponding to the training goals, including specific evaluation indicators in four dimensions: training execution evaluation, functional improvement evaluation, symptom control evaluation, and safety monitoring evaluation.

[0060] Preferably, a multi-perspective time series association analysis method is applied to analyze the mutual influence relationship and time series evolution pattern among the evaluation indicators, and establish an indicator association map, including the following steps: preprocessing the time series data collected during the patient rehabilitation training; based on the sliding time window technology, dividing the preprocessed time series data into multiple overlapping time segments, and extracting the statistical characteristics of the evaluation indicators in each time segment; constructing a multidimensional indicator association network based on a directed graph, in which the nodes represent the evaluation indicators and the edges represent the association relationship between the indicators; applying the conditional mutual information algorithm to calculate the mutual information value between the evaluation indicators in different time windows, wherein the indicator value of the historical time window is used as the conditional variable to determine the time series dependency relationship between the indicators; verifying the causal relationship between the indicators through the Granger causality test method, combining the mutual information value and the p-value of the Granger causality test to calculate the comprehensive association strength value as the edge weight of the directed graph, and when the results of the two methods are inconsistent, the Granger causality test result is preferably used; based on the community detection algorithm, identifying the related indicator groups, and annotating them in combination with clinical knowledge to form the final indicator association map. By constructing an indicator correlation map, the dynamic correlation relationship and temporal evolution characteristics between evaluation indicators are revealed, providing a data support basis for subsequent evaluation and analysis, and helping to more accurately understand the mutual influence mechanism between various indicators.

[0061] Furthermore, a hierarchical Bayesian model was applied to integrate the indicator association map with the patient's historical rehabilitation data to determine the weight coefficients of each evaluation indicator at different rehabilitation stages. The model included the following steps: Based on the indicator association map, a hierarchical model of evaluation indicators was constructed, organizing the evaluation indicators of the four dimensions into a two-level hierarchy: a top-level dimension layer and a bottom-level specific indicator layer; The patient's historical rehabilitation data was annotated by stage based on the clinical rehabilitation process; For each rehabilitation stage, the correspondence between the evaluation indicators and the rehabilitation effect in the patient's historical data was extracted to form a training dataset, where the rehabilitation effect was quantified based on the magnitude of improvement on a standard functional assessment scale; A prior distribution for the hierarchical Bayesian model was constructed, and the prior probability distribution of the model parameters was initialized based on the correlation strength in the indicator association map; The Markov Chain Monte Carlo method was used to perform posterior inference on the hierarchical Bayesian model to obtain the posterior sample distribution of each indicator parameter; and the standardized weight coefficients of each evaluation indicator at different rehabilitation stages were calculated based on the expected value of the posterior distribution. By integrating the indicator association map with historical rehabilitation data, dynamic adjustment of the evaluation indicator weights was achieved, enabling the evaluation results to better reflect the characteristics of different rehabilitation stages and improving the targeted nature of the evaluation.

[0062] Furthermore, the patient feedback data is mapped to a multidimensional rehabilitation evaluation index system to generate standardized scores for the corresponding evaluation indicators. This step is achieved by generating standardized scores of 0-100 points through feature extraction and segmented mapping rules.

[0063] Furthermore, a comparative analysis is conducted based on the standardized scores and the phased goals of the training program to calculate the target achievement of each evaluation indicator, wherein the target achievement is determined by calculating the ratio of the standardized score to the target value. The target achievement of each evaluation indicator is weightedly calculated by applying a weight coefficient, and the influence link is analyzed based on the indicator association map to identify priority adjustment indicators, including the following steps: calculating the weighted target achievement of each evaluation indicator, multiplying the target achievement of each indicator by the weight coefficient of the corresponding rehabilitation stage to obtain the weighted target achievement value; calculating the overall rehabilitation progress score, summing the weighted target achievement of all evaluation indicators, and calculating the overall rehabilitation progress score; identifying key gap indicators, and screening out evaluation indicators with target achievement lower than the target achievement threshold and weight coefficient greater than the weight median as candidate adjustment targets. The target achievement threshold is determined based on the expected completion rate of the rehabilitation stage goals and is set to 0.75; based on the indicator association map, constructing an influence propagation tree for each candidate adjustment target to track the potential impact path and influence intensity of the indicator on other indicators; calculating the influence index of each candidate adjustment target, comprehensively considering its weight coefficient, target gap size and diffusion range in the influence propagation tree; sorting the candidate adjustment targets according to the influence index, and selecting the indicator with the highest influence index as the priority adjustment indicator. The analysis method based on goal achievement and influence index enables the system to scientifically identify indicators that need priority adjustment, providing a more targeted basis for adjusting the training program and reducing subjectivity in the evaluation and adjustment process.

[0064] Furthermore, based on the priority adjustment indicators, a rehabilitation training program adjustment strategy is generated to adjust the rehabilitation training program.

[0065] Optimally, by establishing a multidimensional rehabilitation evaluation indicator system and integrating temporal correlation analysis with a hierarchical Bayesian model, dynamic evaluation and precise adjustment of rehabilitation training programs are achieved. This technical solution accurately identifies key indicators that influence rehabilitation outcomes and systematically analyzes the correlations between indicators and their temporal evolutionary characteristics, enabling scientific adjustments to training programs. This effectively addresses the issues of fragmented evaluation indicators and subjective adjustment strategies during rehabilitation training.

[0066] The processing flow of the incremental learning update module includes: calculating the deviation index between the patient's actual rehabilitation execution results and the predicted rehabilitation process, and determining the main characteristic dimensions of the prediction deviation, specifically including: constructing a time series sequence of the patient's actual rehabilitation trajectory, and forming a set of actual rehabilitation execution results based on the evaluation index data collected regularly; obtaining the rehabilitation process prediction results of the feature association prediction model at the corresponding time point; calculating the time series deviation value of each evaluation indicator; statistically analyzing the distribution characteristics of each deviation indicator and calculating the deviation significance level; according to the deviation significance analysis results, determining the characteristic dimension with significant difference in prediction effect as the main characteristic dimension.

[0067] Furthermore, the model features and parameters that need to be optimized are determined based on the deviation index analysis, including the following steps: constructing a feature-parameter mapping relationship table to establish the correspondence between the main feature dimensions and the model parameters; performing parameter sensitivity analysis to evaluate the impact of perturbed parameter values ​​on the prediction results; combining the deviation analysis results and parameter sensitivity to determine the features and parameters that need to be optimized; and formulating a parameter adjustment strategy based on the deviation direction and magnitude of the feature dimensions.

[0068] Furthermore, historical data related to the main feature dimensions are extracted from the patient rehabilitation database to construct an incremental learning dataset; the parameters of the feature association prediction model are updated, and the model parameters related to the main feature dimensions are adjusted; the updated feature association prediction model is verified to evaluate the model prediction accuracy and performance improvement.

[0069] Furthermore, this embodiment also provides an intelligent analysis method for rehabilitation treatment of lumbar disc herniation, including: obtaining multimodal medical data of patients with lumbar disc herniation; preprocessing the multimodal medical data, applying a feature importance adaptive ranking algorithm to screen a rehabilitation prediction feature set, and dynamically optimizing feature weights based on the patient's rehabilitation stage; constructing a feature association prediction model, inputting the rehabilitation prediction feature set into the feature association prediction model, and obtaining the patient's rehabilitation process prediction results; based on the rehabilitation process prediction results, performing action combination selection and training parameter optimization through a multi-objective optimization algorithm to generate a rehabilitation training plan; monitoring the feedback data of the patient during the execution of the digital rehabilitation plan; constructing a multidimensional rehabilitation evaluation index system, evaluating the rehabilitation process status based on the feedback data, and dynamically adjusting the rehabilitation training plan according to the evaluation results; based on the difference between the execution results of the adjusted rehabilitation training plan and the rehabilitation process prediction results, updating the feature association prediction model through incremental learning.

[0070] In summary, the present invention can realize accurate customization of rehabilitation programs according to individual characteristics of patients and characteristics of rehabilitation stages through feature adaptive optimization and patient classification prediction, improve the matching degree between training programs and patient status, and provide strong support for personalized rehabilitation treatment; based on the multidimensional rehabilitation evaluation index system and feedback data, a real-time monitoring and dynamic adjustment mechanism of the rehabilitation process is established, which can timely identify deviations in rehabilitation effects, dynamically optimize training programs, and ensure the continued effectiveness of rehabilitation treatment; through comprehensive analysis of multimodal medical data and continuous optimization of models, the subjectivity in the process of rehabilitation program formulation is reduced, and reliable decision-making support is provided for clinical rehabilitation treatment; combined with the risk level assessment and dynamic monitoring mechanism of training movements, the safety control of training programs is achieved, potential risks can be timely identified and avoided, and the safety of the rehabilitation training process is ensured.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent analysis system for rehabilitation treatment of lumbar disc herniation, characterized in that: include: Multimodal data acquisition module, used to obtain multimodal medical data of patients with lumbar disc herniation; The feature adaptive optimization module is used to preprocess multimodal medical data, apply a feature importance adaptive ranking algorithm to screen the recovery prediction feature set, and dynamically optimize feature weights based on the patient's recovery stage; The rehabilitation prediction analysis module is used to analyze the rehabilitation prediction feature set based on the feature association prediction model and generate the patient's rehabilitation process prediction results; A training program generation module is used to generate a rehabilitation training program based on the rehabilitation process prediction results through a multi-objective optimization algorithm; Feedback data collection module, used to monitor the feedback data of patients during the implementation of rehabilitation training programs; The rehabilitation assessment and adjustment module is used to evaluate the status of the rehabilitation process based on the multi-dimensional rehabilitation evaluation indicator system and feedback data, and dynamically adjust the rehabilitation training plan according to the assessment results; The incremental learning update module is used to update the feature association prediction model through incremental learning based on the difference between the execution results of the adjusted rehabilitation training program and the predicted results of the rehabilitation process.

2. The intelligent analysis system for rehabilitation treatment of lumbar disc herniation according to claim 1, characterized in that: The multimodal medical data includes clinical assessment data, imaging data, rehabilitation training data and daily activity monitoring data.

3. The intelligent analysis system for rehabilitation treatment of lumbar disc herniation according to claim 1, characterized in that: The processing flow of the feature adaptive optimization module includes: Extracting candidate feature sets from preprocessed multimodal medical data; A combined feature evaluation method is used to calculate the initial importance score of each candidate feature and generate a feature priority list; Construct feature correlation matrix, calculate mutual information and correlation coefficient between features, and identify redundant features and complementary features; Based on the feature correlation analysis results, the feature priority list is adjusted for correlation penalties to establish a feature benchmark ranking; Establish a stage feature mapping relationship based on the patient's recovery stage and dynamically adjust the feature weight coefficient; Based on the adjusted feature weights and feature benchmark rankings, the optimal feature subset is selected from the candidate feature set to generate the rehabilitation prediction feature set.

4. The intelligent analysis system for rehabilitation treatment of lumbar disc herniation according to claim 1, characterized in that: The processing flow of the rehabilitation prediction and analysis module includes: Design the infrastructure for feature association prediction models; A multidimensional clinical feature classification strategy was used to divide patients into multiple disc herniation subgroups; Based on a historical patient rehabilitation database, the feature association prediction model is pre-trained through transfer learning, and the Bayesian optimization method is used to automatically select the optimal model parameters for each subgroup of patients; Design a multi-objective joint learning framework to simultaneously train three prediction tasks: functional recovery, symptom relief, and relapse risk, sharing the underlying feature representation; Evaluate model performance and optimize the model's predictive calibration using Bayesian posterior probability calibration techniques combined with clinical decision thresholds; Output the patient's recovery process prediction results.

5. The intelligent analysis system for rehabilitation treatment of lumbar disc herniation according to claim 1, characterized in that: The processing flow of the training scheme generation module includes: Build a rehabilitation training action library and establish a dual attribute mapping including rehabilitation goals and risk levels for each training action; Construct a multi-objective optimization function and formulate phased rehabilitation goals and evaluation criteria; Determining a training parameter limit range based on the staged rehabilitation goals and rehabilitation progress prediction results, and selecting candidate action combinations from a rehabilitation training action library using a risk-constrained hierarchical search strategy; Applying the analytic hierarchy process to prioritize the candidate action combinations, generating the optimal action sequence, and using Bayesian optimization to determine the rehabilitation training prescription parameters for each training action; Integrate the optimal movement sequence, training prescription parameters and phased rehabilitation goals to generate a rehabilitation training plan.

6. The intelligent analysis system for rehabilitation treatment of lumbar disc herniation according to claim 1, characterized in that: The processing flow of the rehabilitation assessment and adjustment module includes: Construct a multidimensional rehabilitation evaluation indicator system; Apply multi-perspective temporal correlation analysis methods to analyze the mutual influence and temporal evolution patterns among evaluation indicators and establish an indicator correlation map; A hierarchical Bayesian model was used to integrate the indicator association map with the patient's historical rehabilitation data to determine the weight coefficients of each evaluation indicator at different rehabilitation stages; Mapping the patient feedback data to the multidimensional rehabilitation evaluation index system to generate standardized scores for the corresponding evaluation indicators; Compare and analyze the standardized scores with the phased goals of the training program and calculate the target achievement of each evaluation indicator; The target achievement degree of each evaluation indicator is weightedly calculated by applying the weight coefficient, and the impact link is analyzed based on the indicator association map to identify the priority adjustment indicator.

7. The intelligent analysis system for rehabilitation treatment of lumbar disc herniation according to claim 1, characterized in that: The processing flow of the incremental learning update module includes: Calculate the deviation index between the patient's actual rehabilitation execution results and the predicted rehabilitation process, and determine the main characteristic dimensions of the prediction deviation; Determining the model features and parameters that need to be optimized based on the deviation index analysis; Extracting historical data related to the main feature dimensions from a patient rehabilitation database to construct an incremental learning dataset; The parameters of the feature association prediction model are updated to adjust the model parameters related to the main feature dimensions.

8. An intelligent analysis method for rehabilitation treatment of lumbar disc herniation, characterized in that: include: Acquire multimodal medical data of patients with lumbar disc herniation; Preprocessing the multimodal medical data, applying a feature importance adaptive ranking algorithm to screen a rehabilitation prediction feature set, and dynamically optimizing feature weights based on the patient's rehabilitation stage; Constructing a feature association prediction model, inputting the rehabilitation prediction feature set into the feature association prediction model, and obtaining a patient's rehabilitation process prediction result; Based on the rehabilitation progress prediction results, performing action combination selection and training parameter optimization through a multi-objective optimization algorithm to generate a rehabilitation training plan; monitoring feedback data from the patient during the process of executing the digital rehabilitation program; Constructing a multidimensional rehabilitation evaluation indicator system, evaluating the rehabilitation process status based on the feedback data, and dynamically adjusting the rehabilitation training program according to the evaluation results; Based on the difference between the execution result of the adjusted rehabilitation training program and the rehabilitation progress prediction result, the feature association prediction model is updated through incremental learning.

Images