Neurology patient rehabilitation nursing method based on multi-modal data analysis

Abstract

The invention discloses a neurology patient rehabilitation nursing method based on multi-modal data analysis, and relates to the technical field of medical health, and the method comprises the steps: extracting neural function features through a multi-modal data fusion algorithm, and carrying out the calculation through a weighted fusion and statistical analysis method, and obtaining a neuroplasticity index vector; combining the neuroplasticity index vector with the unified multi-modal feature representation, carrying out multi-modal abnormal mode recognition analysis, obtaining a rehabilitation risk early warning signal and a personalized intervention suggestion, and generating a personalized rehabilitation scheme; performing dynamic optimization and self-adaptive adjustment on the individualized rehabilitation scheme by using a feedback loop mechanism to generate an optimized individualized rehabilitation scheme; physiological signals in individualized rehabilitation training based on the optimized individualized rehabilitation scheme are collected in real time and preprocessed, and a standardized physiological response data sequence is generated. According to the invention, the scientificity, timeliness and individual adaptability of regulation and control are improved, so that nerve function remodeling is accelerated and the rehabilitation risk is reduced.

Description

Technical Field

[0001] The present invention relates to the field of medical health technology, and in particular to a rehabilitation nursing method for neurology patients based on multimodal data analysis. Background Art

[0002] With the increasing prevalence and chronicity of neurological diseases, clinical rehabilitation nursing is evolving from simple basic care to refined, data-driven rehabilitation interventions. During the rehabilitation process for typical conditions such as stroke, Parkinson's disease, and Alzheimer's disease, timely and accurate assessment of the patient's neurological status, guiding individualized interventions, is crucial for improving rehabilitation outcomes. Common rehabilitation nursing assessment methods currently include scales (such as the Fugl-Meyer and NIHSS), subjective physician assessments, single-point EEG / EMG testing, and imaging studies. While these methods offer valuable insights in specific scenarios, they generally rely on intermittent data collection and static analysis, failing to continuously reflect dynamic changes during the rehabilitation process. Often based on a single modality or a small number of indicators, they struggle to fully capture the multidimensional characteristics of a patient's neurological status, leading to delayed responses and interventions.

[0003] Patients' physiological states exhibit significant individual differences and dynamic fluctuations. Failure to capture these changes and adjust intervention content in a timely manner can easily lead to mismatched intervention rhythms, diminished training effectiveness, and even the occurrence of risk events. Current conventional approaches fail to effectively leverage the latent structural information in time series for real-time analysis of the rehabilitation process, nor to establish intervention optimization strategies based on data feedback, hindering the development of intelligent and precise neurorehabilitation care. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a rehabilitation nursing method for neurological patients based on multimodal data analysis to solve the problem of lack of temporal dynamic modeling capability of multimodal data of physiological state.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a rehabilitation nursing method for neurological patients based on multimodal data analysis, which includes: collecting multimodal data of neurological patients, and performing standardized preprocessing to form a unified multimodal feature representation; extracting neural function characteristics through a multimodal data fusion algorithm and calculating them using weighted fusion and statistical analysis methods to obtain a neural plasticity index vector; combining the neural plasticity index vector with the unified multimodal feature representation, performing multimodal abnormal pattern recognition analysis, obtaining rehabilitation risk warning signals and personalized intervention suggestions, and generating an individualized rehabilitation plan; using a feedback loop mechanism to dynamically optimize and adaptively adjust the individualized rehabilitation plan to generate an optimized individualized rehabilitation plan; real-time collection of physiological signals in individualized rehabilitation training based on the optimized individualized rehabilitation plan, and preprocessing to generate a standardized physiological response data sequence; using time series analysis and multimodal feature fusion methods to dynamically model the standardized physiological response data sequence, and analyze abnormal states during rehabilitation training.

[0007] As a preferred solution of the rehabilitation nursing method for neurological patients based on multimodal data analysis of the present invention, the specific steps of forming a unified multimodal feature representation are as follows: Collect EEG, imaging, physiology, behavior and questionnaires from neurology patients to form multimodal data; Anomaly detection and interpolation repair methods are used to clean multimodal data and generate cleaned multimodal datasets; The cleaned multimodal dataset is format converted and time-series aligned to construct a structured time-synchronized multimodal data sample. Normalization and Z-score standardization methods are used to process it and generate standardized multimodal feature data. The feature mapping function is used to unify the dimensions and spatially project the standardized multimodal feature data to form a unified multimodal feature representation.

[0008] As a preferred solution of the rehabilitation nursing method for neurological patients based on multimodal data analysis of the present invention, the steps of obtaining the neuroplasticity index vector are as follows: A feature screening algorithm is used to reduce the dimensionality of the unified multimodal feature representation, extract a subset of candidate features related to neural function, and screen out a set of highly relevant neural function features through correlation analysis and importance assessment. A weighted fusion method is used to fuse heterogeneous modal features in a set of highly correlated neural function features to construct a neural function fusion vector. Statistical analysis methods were used to perform attribution modeling and index calculation on the neural function fusion vector to obtain the neural plasticity index vector.

[0009] As a preferred solution of the rehabilitation nursing method for neurological patients based on multimodal data analysis of the present invention, the specific steps of generating an individualized rehabilitation plan are as follows: The neuroplasticity index vector is fused with the unified multimodal feature representation to form a fused feature representation, and statistical analysis methods are used to detect abnormal patterns, identify abnormal features and generate rehabilitation risk warning signals; Based on the rehabilitation risk warning signals and combined with the individual characteristics of patients in the unified multimodal feature representation, personalized intervention recommendations are formulated and an individualized rehabilitation plan is generated.

[0010] As a preferred solution of the rehabilitation nursing method for neurological patients based on multimodal data analysis of the present invention, the specific steps of generating an optimized individualized rehabilitation plan are as follows: Collect real-time physiological response indicators during individualized rehabilitation training, use bandpass filtering to preprocess the signals, and construct a rehabilitation response data sequence; Using time series feature extraction algorithms and long short-term memory networks, dynamic feature extraction and trend modeling are performed on the rehabilitation response data sequence to generate individual rehabilitation status assessment results; Based on the differences between individual rehabilitation status assessment results and individual rehabilitation target parameters, the adaptive deviation of the individual rehabilitation program is analyzed, and an optimization algorithm is used to generate program adjustment suggestions; According to the program adjustment suggestions, the rehabilitation control parameters in the individualized rehabilitation program are updated, and the structure is optimized to generate an optimized individualized rehabilitation program.

[0011] As a preferred embodiment of the rehabilitation nursing method for neurological patients based on multimodal data analysis described in the present invention, the adaptive deviation of the individualized rehabilitation program is analyzed according to the difference between the individual rehabilitation status assessment results and the individualized rehabilitation target parameters. The specific steps are as follows: Based on multimodal data, clinical information extraction and feature mapping methods are used to extract patients' diagnostic labels, functional status scores, and treatment plan elements, and construct individualized rehabilitation target parameters. Using structured extraction method, individualized rehabilitation goal parameters are structured and standardized rehabilitation goal vectors are generated. Based on the individual rehabilitation status assessment results, the rehabilitation response data sequence is analyzed using the time series regression modeling method to extract dynamic features and generate the current rehabilitation status feature vector; The current rehabilitation state feature vector is compared with the standardized rehabilitation target vector to obtain the index difference value, and the deviation analysis algorithm is used to quantify the degree of adaptive deviation between the individualized rehabilitation plan and the individualized rehabilitation target parameters.

[0012] As a preferred solution of the rehabilitation nursing method for neurological patients based on multimodal data analysis described in the present invention, it comprises: formulating a personalized rehabilitation training plan and starting rehabilitation training based on the optimized individualized rehabilitation plan, collecting individual physiological signal data in real time during the training process, removing noise and unifying dimensions, and generating a standardized physiological response data sequence.

[0013] As a preferred solution of the rehabilitation nursing method for neurological patients based on multimodal data analysis described in the present invention, the method utilizes time series analysis and multimodal feature fusion method to dynamically model the standardized physiological response data sequence and analyze abnormal states during rehabilitation training. The specific steps are as follows: Using time series analysis methods, the standardized physiological response data series is dynamically modeled, the time-dependent characteristics of the physiological response are extracted, and the physiological dynamic change characteristics are obtained; Using multimodal feature fusion methods, the physiological dynamic change characteristics are fused with the rehabilitation response data sequence to generate a real-time physiological state feature vector; Perform pattern recognition and trend evaluation on real-time physiological state feature vectors, analyze abnormalities in state changes during rehabilitation training, and dynamically adjust the optimized individualized rehabilitation plan when abnormalities are detected.

[0014] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the rehabilitation nursing method for neurological patients based on multimodal data analysis as described in the first aspect of the present invention is implemented.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the rehabilitation care method for neurological patients based on multimodal data analysis as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are: by deeply fusing the neuroplasticity index vector with the standardized unified multimodal feature representation, and combining it with statistical analysis methods to detect abnormal patterns, it achieves accurate identification of potential risks in the rehabilitation process; further generates personalized intervention suggestions and rehabilitation plans based on individual characteristics, effectively opens up the dynamic linkage mechanism between changes in neural function and behavioral feedback, breaks through the limitations of traditional reliance on a single modality and experience-based evaluation, realizes the transformation of the rehabilitation process from passive response to active prediction, and significantly improves the scientificity, timeliness and individual adaptability of the intervention, thereby accelerating neural function remodeling and reducing rehabilitation risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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.

[0018] Figure 1 This is a flow chart of the rehabilitation nursing method for neurological patients based on multimodal data analysis in the present invention.

[0019] Figure 2 Flowchart of unified multimodal feature representation in the present invention.

[0020] Figure 3 Flowchart generated for the individualized rehabilitation program in the present invention.

[0021] Figure 4 This is a flow chart for the dynamic optimization of the individualized rehabilitation program in the present invention. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a rehabilitation nursing method for neurological patients based on multimodal data analysis, comprising the following steps: S1. Collect multimodal data of neurology patients and perform standardized preprocessing to form a unified multimodal feature representation.

[0026] S1.1. Collect EEG, imaging, physiology, behavior and questionnaire data from neurology patients to form multimodal data.

[0027] Specifically, multi-channel EEG acquisition equipment was used to synchronously acquire EEG signals from neurological patients in resting and task states, with an EEG acquisition frequency of 1000 Hz and an acquisition time of 10 minutes per session. Structural magnetic resonance imaging and functional magnetic resonance imaging equipment were used to acquire brain imaging data from neurological patients, including three-dimensional T1-weighted images and resting-state fMRI images. An example of image acquisition parameters was a resolution of 1×1×1 mm³, and an example of a scanning time of 8 minutes per session. Multi-channel physiological monitoring equipment was used to record physiological data such as heart rate, skin conductance, and respiratory rate from neurological patients, with an example sampling frequency of 500 Hz and acquisition time synchronized with EEG. Behavioral recording devices were used to collect behavioral data such as movement trajectory, movement amplitude, and reaction time from neurological patients during standardized tasks, with an example of a grasping movement tracking task. Five structured questionnaires were completed in paper or electronic form before and after the intervention, covering cognitive state, emotional state, and daily functioning level. Each questionnaire contained 30 scoring items, and a five-level scale was used for assessment. EEG, imaging, physiological, behavioral and questionnaire data are uniformly time-stamped and archived into a multimodal data set.

[0028] S1.2. Use anomaly detection and interpolation repair methods to clean the multimodal data and generate a cleaned multimodal dataset.

[0029] Specifically, format compliance checks and timestamp consistency checks were performed on five data types: EEG, imaging, physiological, behavioral, and questionnaire data. Data entries with inconsistent timestamps or missing fields were removed. Anomaly detection was performed on the EEG and physiological data using a Z-score-based method, with a threshold of ±3 (for example). Data points with amplitudes exceeding the Z-score threshold were marked. Linear interpolation was used to repair data segments marked as anomalies in the EEG and physiological data, with a repair range of no more than 5 seconds of consecutive missing points. For behavioral data, the coefficient of variation of velocity and acceleration was calculated to identify mutation points, and spline interpolation was applied to these mutation segments. For imaging data, a structural consistency check was used to exclude image slices with motion artifacts or reconstruction failures, and unqualified images were deleted. For questionnaire data, logical consistency rules were used to identify internal inconsistencies, and individual missing items were filled using mean interpolation, with the proportion of missing items controlled to less than 5%. The processed EEG, imaging, physiological, behavioral, and questionnaire data were stored as a cleaned multimodal dataset.

[0030] It should also be explained that the specific steps for setting the Z-score anomaly detection threshold are: obtaining the original data sequence of EEG data and physiological data within a certain time window; calculating the mean and standard deviation of the EEG data and physiological data respectively; standardizing each data point according to the mean and standard deviation; setting the Z-score anomaly detection threshold to ±3, identifying all EEG data points and physiological data points with an absolute value greater than 3 after standardization, and marking them as abnormal data points; recording the marking results and the time series position of the abnormal data points to analyze the state changes during rehabilitation training and assist in dynamically adjusting the optimized individualized rehabilitation plan.

[0031] S1.3. Perform format conversion and time alignment on the cleaned multimodal dataset to construct a structured time-synchronized multimodal data sample. Use normalization and Z-score standardization methods to process it and generate standardized multimodal feature data.

[0032] Specifically, the EEG data, physiological data, behavioral data, and questionnaire data in the cleaned multimodal dataset are uniformly converted into a data table structure arranged by time index, with each record containing a timestamp, data type, numerical field, and source information. Using the EEG data timestamp as the primary time axis, the physiological data and behavioral data are time-series aligned using linear interpolation, for example, every 100 milliseconds. For the questionnaire data, static label expansion is used to flatten the once-acquired data over time and fill it into the entire observation period. For the imaging data, the nearest time point is matched based on the image acquisition timestamp and the EEG data timestamp, and the corresponding imaging identifier is attached to the data record of the corresponding time period. After completing these steps, a structured time-synchronized multimodal data sample is formed. Each feature type in the structured time-synchronized multimodal data sample is subjected to normalization and Z-score standardization. Normalization is used to scale the feature values ​​to a uniform numerical range, and Z-score standardization is used to eliminate scale differences between different features, ultimately generating standardized multimodal feature data.

[0033] S1.4. Use the feature mapping function to unify the dimensions and spatially project the standardized multimodal feature data to form a unified multimodal feature representation.

[0034] Specifically, the dimensional information of EEG feature data, physiological feature data, behavioral feature data, imaging feature data and questionnaire feature data in the statistically standardized multimodal feature data is collected; the principal component analysis method is used for the EEG feature data, by calculating the covariance matrix, extracting the principal component vector and projecting it to the first several principal component spaces, for example, the first 20 principal components are selected as the mapped EEG feature representation; the linear discriminant analysis method is used for the behavioral feature data, by calculating the between-class scatter matrix and the within-class scatter matrix and solving the generalized eigenvalues, extracting the eigenvectors with discriminant ability for transformation; the t-SNE method is used for the imaging feature data, first performing high-dimensional similarity calculation, and then optimizing the distribution structure of the low-dimensional embedding to retain local information; the maximum variance retention strategy is adopted for the physiological feature data and the questionnaire feature data, respectively, and the first several dimensional features are selected for projection through feature covariance analysis; finally, the processed EEG feature representation, physiological feature representation, behavioral feature representation, imaging feature representation and questionnaire feature representation are spliced ​​and combined according to a unified dimension to generate a unified multimodal feature representation.

[0035] S2. Through the multimodal data fusion algorithm, the neural function characteristics are extracted and calculated using weighted fusion and statistical analysis methods to obtain the neural plasticity index vector.

[0036] S2.1. Use feature screening algorithm to reduce the dimensionality of unified multimodal feature representation, extract candidate feature subsets related to neural function, and screen out highly correlated neural function feature sets through correlation analysis and importance evaluation.

[0037] Specifically, the Pearson correlation coefficient between each dimension of the feature in the unified multimodal feature representation and the neurological function score is calculated respectively, and the features with the absolute value of the correlation coefficient greater than, for example, 0.4 are screened out to form the initial candidate feature subset; the variance selection method is used for the initial candidate feature subset to eliminate low-variance features with variance less than, for example, 0.01; the extreme gradient boosting method is used to train the initial candidate feature subset, and the output initial candidate feature importance score is sorted from high to low according to importance; the first several features with importance scores higher than the preset random forest feature importance score threshold are selected as the high-correlation neurological function feature set.

[0038] The specific steps for presetting the random forest feature importance score threshold should also be explained: based on historical experimental data, statistically analyze the distribution characteristics of the random forest feature importance scores in multiple sample data; analyze the distribution curve of the feature importance scores, and determine a critical value that can effectively distinguish important and unimportant features as a reference; verify the screening effect of the random forest feature importance score threshold on different data sets through cross-validation methods to ensure that features highly correlated with neural function can be stably selected; combined with specific application requirements and comprehensive consideration of the number of features, determine the final random forest feature importance score threshold, for example, set it to 0.02, for selecting a set of highly correlated neural function features in subsequent screening steps.

[0039] S2.2. Use the weighted fusion method to fuse the heterogeneous modal features in the highly correlated neural function feature set and construct a neural function fusion vector.

[0040] Specifically, the set of highly correlated neural function features is classified according to the modality type to obtain EEG modal features, imaging modal features, physiological modal features, behavioral modal features and questionnaire modal features; the random forest feature importance score of each of the EEG modal features, imaging modal features, physiological modal features, behavioral modal features and questionnaire modal features is calculated respectively, and the importance scores of all features in the same modality are normalized as the weight coefficient of the random forest feature; each heterogeneous modal feature is multiplied by the corresponding weight coefficient to obtain a weighted eigenvalue; the weighted eigenvalues ​​of different modalities are then added element by element according to the feature dimension to obtain the comprehensive feature vector before fusion; the comprehensive feature vector before fusion is processed by the maximum and minimum normalization method so that all values ​​fall between 0 and 1 to generate a neural function fusion vector.

[0041] S2.3. Use statistical analysis methods to perform attribution modeling and index calculation on the neural function fusion vector to obtain the neural plasticity index vector.

[0042] Specifically, the least squares method is used to perform linear regression on each dimension of heterogeneous modal features in the neural function fusion vector and the behavioral variables, and the regression coefficient of each dimension of heterogeneous modal features is calculated as the attribution weight; each dimension of heterogeneous modal features in the neural function fusion vector is multiplied by the corresponding attribution weight and then summed to obtain the initial neural plasticity score value of each sample; the mean and standard deviation of the initial neural plasticity score values ​​are calculated, and the Z-score normalization method is used to subtract the mean from each initial neural plasticity score value and divide it by the standard deviation to form a standardized score sequence; the standardized score sequence is output as the neural plasticity indicator vector.

[0043] S3. Combine the neuroplasticity index vector with the unified multimodal feature representation to perform multimodal abnormal pattern recognition analysis, obtain rehabilitation risk warning signals and personalized intervention recommendations, and generate an individualized rehabilitation plan.

[0044] S3.1. Fuse the neuroplasticity index vector with the unified multimodal feature representation to form a fused feature representation, and use statistical analysis methods to detect abnormal patterns, identify abnormal features, and generate rehabilitation risk warning signals.

[0045] Specifically, the neuroplasticity index vector and the unified multimodal feature representation are spliced ​​according to the feature dimension to generate a fused feature representation; the mean and standard deviation of each dimension of the fused feature representation are calculated; the Z-score normalization method is applied to each dimension of the fused feature representation of each sample, and the standardized value is calculated; then, according to the set abnormal detection threshold (for example, the absolute value of the Z-score is greater than 3), the corresponding dimension is marked as an abnormal feature; if the number of marked abnormal features in a single sample exceeds the set abnormal feature number threshold (for example, more than 5), a rehabilitation risk warning signal is generated.

[0046] It should also be explained that the specific steps for setting the anomaly detection threshold are: for each dimension in the fusion feature representation, first count the numerical distribution of all samples in each dimension, including the center position and fluctuation range, then calculate the degree of deviation between the value of the current sample to be tested in each dimension and the overall value, and set the anomaly detection threshold (for example, the degree of deviation exceeds three times the normal range). If the degree of deviation of the value of a certain dimension exceeds the anomaly detection threshold, the corresponding fusion feature is marked as an abnormal feature.

[0047] The specific steps for setting the threshold for the number of abnormal features are as follows: after completing the abnormal marking for each dimension represented by the fusion feature, count the number of fusion feature dimensions marked as abnormal in the current sample to be tested, and set the threshold for the number of abnormal features (for example, the number reaches or exceeds 5). If the number of fusion feature dimensions marked as abnormal is greater than or equal to the threshold for the number of abnormal features, it is identified as an overall abnormality, and a rehabilitation risk warning signal is generated.

[0048] S3.2. Based on the rehabilitation risk warning signals and combined with the individual characteristics of patients in the unified multimodal feature representation, formulate personalized intervention recommendations and generate individualized rehabilitation plans.

[0049] Specifically, individual patient characteristics, including physiological parameters, behavioral characteristics, and historical rehabilitation data, are extracted from the unified multimodal feature representation. Subsequently, the individual patient characteristics are matched with rehabilitation risk warning signals and analyzed. Intervention measures corresponding to the patient's risk characteristics are screened out based on a pre-established intervention strategy library. Finally, based on the screened intervention measures and the weights of the patient's individual characteristics, an individualized rehabilitation plan is comprehensively generated and output for subsequent implementation.

[0050] The specific steps for establishing an intervention strategy library in advance should also be explained: collect a large amount of rehabilitation clinical data and organize them into a list of intervention measures; classify and manage the intervention measures, and label them according to rehabilitation goals, applicable symptoms and patient characteristics; associate each intervention measure with the corresponding patient characteristics and rehabilitation risk types to form strategy matching rules; finally, enter all intervention measures and matching rules into the database to complete the construction of the intervention strategy library.

[0051] S4. Use the feedback loop mechanism to dynamically optimize and adaptively adjust the individualized rehabilitation plan to generate an optimized individualized rehabilitation plan.

[0052] S4.1. Collect real-time physiological response indicators during individualized rehabilitation program training, use bandpass filtering method to preprocess the signals, and construct a rehabilitation response data sequence.

[0053] Specifically, real-time physiological response indicators, including heart rate, galvanic skin response, etc., are collected during the training of the individualized rehabilitation program. The original physiological signals are obtained through the acquisition equipment, and the original physiological signals are preprocessed using the bandpass filtering method. The low-frequency cutoff frequency and high-frequency cutoff frequency of the bandpass filter are set, for example, the low-frequency cutoff frequency is 0.5 Hz, and the high-frequency cutoff frequency is 40 Hz. The collected original physiological signals are filtered to remove noise and interference to obtain filtered physiological response signals. The filtered physiological response signals are sampled and stored in chronological order to form a continuous rehabilitation response data sequence.

[0054] It should also be explained that the specific steps for setting the low-frequency cutoff frequency and high-frequency cutoff frequency of the band-pass filter are as follows: determine the analyzable frequency range based on the sampling rate of the EEG data and physiological data; select the target frequency band based on the physiological signal characteristics of interest, for example, if the EEG data focuses on alpha waves, select a low-frequency cutoff frequency of 8 Hz and a high-frequency cutoff frequency of 13 Hz; then set the filter order and filter type of the band-pass filter, for example, select a finite impulse response filter and set the filter order to 64; apply the band-pass filter to filter the EEG data and physiological data, retaining only the signal components between the low-frequency cutoff frequency and the high-frequency cutoff frequency; output the filtered data sequence for subsequent processing and analysis.

[0055] S4.2. Use the time series feature extraction algorithm and long short-term memory network to perform dynamic feature extraction and trend modeling on the rehabilitation response data sequence to generate individual rehabilitation status assessment results.

[0056] Specifically, a time series feature extraction algorithm is used to divide the rehabilitation response data sequence into windows based on a sliding window of fixed time length, for example, every 10 seconds is an analysis cycle, and time series statistical features including mean, standard deviation, slope, kurtosis, etc. are extracted in each window to form a time series feature vector; the time series feature vector is input into the long short-term memory network in chronological order, and the time dependency and evolution trend are obtained through the training phase, and the predicted state value of each time point is output based on the long short-term memory network when a new sequence is input; the output predicted state values ​​are summarized and analyzed, and combined with the evaluation labels in the rehabilitation response data sequence to generate individual rehabilitation status evaluation results.

[0057] It should also be explained that the specific steps for inputting the time series feature vectors into the long short-term memory network in chronological order and obtaining the time dependency and evolution trend through the training stage are as follows: collect and organize the time series feature vectors arranged in chronological order, and divide them into a training set and a validation set, for example, divide them in a ratio of 7:3; normalize the time series feature vectors to keep their numerical range consistent; then, set the input dimension, number of hidden layer units, time step length and training rounds of the long short-term memory network, for example, the input dimension is 10, the number of hidden units is 128, the time step is 20, and the training rounds are 100; then select the loss function and optimization algorithm, for example, use the mean square error as the loss function and the Adam algorithm as the optimization algorithm; then use the training set data to iteratively train the long short-term memory network, and update the network parameters through forward propagation and back propagation in each round; use the validation set to evaluate the training effect, monitor the trend of the loss value change, and decide to stop the training early.

[0058] S4.3. Based on the differences between the individual rehabilitation status assessment results and the individualized rehabilitation target parameters, analyze the adaptive deviation of the individualized rehabilitation program and use the optimization algorithm to generate program adjustment suggestions.

[0059] It should also be explained that the specific process of using the optimization algorithm to generate program adjustment suggestions is as follows: based on the individual rehabilitation status assessment results, the parameters of the currently implemented individualized rehabilitation program are extracted, including training frequency, training intensity, duration, and feedback response indicators as optimization variables; the objective function is to minimize the triggering frequency of the rehabilitation risk warning signal, and the constraints include the maximum training load threshold allowed by individual characteristics, the minimum rest time threshold, and the target functional recovery degree threshold; the particle swarm optimization algorithm is used to iteratively solve the objective function and constraints, and the particle position is represented as a candidate individualized rehabilitation program parameter combination, and the speed update is adjusted according to the current optimal and global optimal states; after the iteration, the parameter combination with the smallest objective function is selected as the program adjustment suggestion to generate the corresponding individualized rehabilitation program parameter update value.

[0060] It should also be noted that the maximum training load threshold, minimum rest time threshold, and target functional recovery threshold allowed by individual characteristics in the constraints are derived from the individual rehabilitation status assessment results and the rehabilitation response data series. The setting process is as follows: Based on individual status characteristics such as current functional level, training response stability index, and fatigue sensitivity index recorded in the individual rehabilitation status assessment results, combined with the maximum training load range in the rehabilitation response data series that did not result in overload reactions or adverse events under similar conditions, a statistical analysis method is used to extract an upper limit value, which is set as the maximum training load threshold allowed by individual characteristics. For example, it is set to an average heart rate per unit time of no more than 140 beats / minute. Combined with the analysis results of the recovery capacity level in the individual rehabilitation status assessment results, the ratio of training frequency to recovery time under similar recovery capacity conditions in the rehabilitation response data series is statistically analyzed to set the minimum rest time threshold, for example, setting the interval between training sessions to no less than 6 hours. Based on the functional recovery target category corresponding to the individual rehabilitation status assessment results, combined with the characteristic change thresholds required to achieve similar recovery goals in the rehabilitation response data series, such as the rate of change in gait cycle stability and the amplitude of myoelectric activity improvement, a target functional recovery threshold is set. For example, it is set to an increase in myoelectric response amplitude of no less than 15%. It is used as a constraint in the dynamic adjustment process of subsequent individual rehabilitation training programs.

[0061] S4.3.1. Based on multimodal data, clinical information extraction and feature mapping methods are used to extract the patient's diagnostic labels, functional status scores, and treatment plan elements, and construct individualized rehabilitation target parameters.

[0062] Specifically, the format of multimodal data including text-formatted electronic medical records, structured examination indicators and image examination results was unified and data cleaned; a named entity recognition method based on bidirectional encoder representation was used to identify clear diagnostic label information from the electronic medical record text, such as disease categories such as "stroke" and "spinal cord injury"; then, functional status scores were extracted from rehabilitation assessment records based on word embedding and rule matching methods, such as "Fugl-Meyer score of 45 points" or "Barthel index of 80 points"; further, keyword annotation and syntactic analysis methods were used to extract treatment plan elements from medical order information and stage treatment descriptions, including training items, training time, training frequency and target improvement areas; finally, the diagnostic labels, functional status scores and treatment plan elements were encoded into structured input variables through a multidimensional feature mapping method to construct individualized rehabilitation target parameters.

[0063] S4.3.2. Use a structured extraction method to perform structured processing on individualized rehabilitation goal parameters and generate a standardized rehabilitation goal vector.

[0064] Specifically, field mapping rules were established for the diagnostic labels, functional status scores, and treatment plan elements in the individualized rehabilitation target parameters. The diagnostic labels were classified and standardized using a label normalization method based on ICD coding, the functional status scores were numerically normalized using an interval division method based on the standard value of the scale, and the treatment plan elements were extracted using a predefined dictionary and rule matching method for fixed-format items. The normalized diagnostic labels were converted into fixed-dimensional vectors using one-hot encoding, the functional status scores were linearly scaled to real values ​​in the range of [0,1] according to the score interval, and each item in the treatment plan element was mapped to a fixed-length vector using position encoding and frequency encoding methods. Finally, the diagnostic label vector, functional status score vector, and treatment plan element vector were sequentially concatenated to generate a standardized rehabilitation target vector with unified dimensions and consistent format.

[0065] It should also be explained that the specific steps for establishing field mapping rules are as follows: based on the three types of fields included in the individualized rehabilitation goal parameters, namely diagnostic labels, functional status scores, and treatment plan elements, the field names and corresponding contents that may appear in the original data are listed respectively; secondly, for diagnostic labels, a correspondence table between diagnostic labels and ICD codes is constructed according to the ICD (International Classification of Diseases) standard to form a normalized mapping relationship between labels; for functional status scores, the type of functional scoring scale used is determined, such as the Barthel index or FIM score, and the original score is divided into fixed intervals according to the scoring criteria, for example, the Barthel index is divided into "severe dependence (0–20)", "moderate dependence (21–60)", "mild dependence (61–90)", "independence (91–100)", etc., and a standardized identifier is assigned to each interval; for treatment plan elements, a dictionary rule table containing elements such as keywords, treatment frequency, and treatment course duration is constructed based on commonly used terms in rehabilitation treatment, and a unified field format and standard value set are defined for each element; finally, the mapping relationship between the three fields is saved in dictionary form for unified field identification and value standardization in subsequent structured processing.

[0066] S4.3.3. Based on the individual rehabilitation status assessment results, use the time series regression modeling method to analyze the rehabilitation response data sequence, extract dynamic features, and generate the current rehabilitation status feature vector.

[0067] Specifically, based on the individual rehabilitation status assessment results, the rehabilitation response data series is time-aligned and missing values ​​are filled to ensure the temporal integrity of the data; a sliding time window is selected, for example, 7 days is used as a time window, and the statistical characteristic indicators in each window are extracted, including mean, standard deviation, maximum value, minimum value, rate of change, etc.; the processed rehabilitation response data series is modeled using the time series regression modeling method, and the ARIMA method or the linear regression method based on the Lasso regular term in the existing time series regression modeling method is selected to perform fitting training on the relationship between the statistical characteristic indicators of each time window and the individual rehabilitation status assessment results; after the training is completed, the rehabilitation response data series in the current time window is input to obtain the output prediction value and the intermediate implicit variable output; finally, all the output intermediate implicit variables are spliced ​​with the prediction results to form the current rehabilitation status feature vector.

[0068] S4.3.4. Compare the current rehabilitation state feature vector with the standardized rehabilitation target vector to obtain the indicator difference value, and use the deviation analysis algorithm to quantify the degree of adaptive deviation between the individualized rehabilitation plan and the individualized rehabilitation target parameters.

[0069] Specifically, the current rehabilitation state feature vector is matched with the standardized rehabilitation target vector one-to-one, and the difference between the corresponding dimension values ​​is calculated to obtain the indicator difference value sequence; based on the established deviation analysis algorithm, the Euclidean distance method or the Mahalanobis distance method is used to quantify the indicator difference value sequence, and the overall deviation between the current rehabilitation state feature vector and the standardized rehabilitation target vector is calculated; then, each indicator difference value is normalized, for example, using the minimum-maximum normalization method to unify each difference value to the [0,1] interval to form a standardized difference vector; the standardized difference vector is weighted and summed using the set indicator weight vector to obtain the adaptive deviation degree value between the individualized rehabilitation plan and the individualized rehabilitation target parameter.

[0070] S4.4. Based on the program adjustment suggestions, update the rehabilitation control parameters in the individualized rehabilitation program, perform structural optimization, and generate an optimized individualized rehabilitation program.

[0071] Specifically, according to the program adjustment suggestions, a set of rehabilitation control parameters in the individualized rehabilitation program is selected, and the parameter values ​​are replaced or the ranges are updated according to the corresponding indicator items in the adjustment suggestions, for example, the rehabilitation frequency is adjusted from 2 times a day to 3 times a day; then, a parameter optimization algorithm, such as a particle swarm optimization method or a genetic algorithm, is used to perform multiple rounds of iterative searches on the updated rehabilitation control parameter set to find the optimal parameter combination under the constraints; then, based on the co-occurrence pattern between the rehabilitation response data sequence and the rehabilitation control parameter set, a statistical correspondence relationship set for optimizing the rehabilitation training parameter combination is extracted, and the arrangement order and execution logic between the rehabilitation control parameters are adjusted to complete the optimization of the rehabilitation control parameter structure; finally, the optimized rehabilitation control parameter set and the structural relationship are output together to generate an optimized individualized rehabilitation program.

[0072] S5. Real-time collection of physiological signals during individualized rehabilitation training based on the optimized individualized rehabilitation program, and preprocessing to generate standardized physiological response data sequences.

[0073] S5.1. Based on the optimized individualized rehabilitation program, formulate a personalized rehabilitation training plan and start rehabilitation training. During the training process, collect individual physiological signal data in real time, remove noise and unify the dimensions, and generate a standardized physiological response data sequence.

[0074] Specifically, based on the optimized individualized rehabilitation plan, a personalized rehabilitation training plan is formulated to clarify the training content, intensity and time arrangement, for example, the daily training duration is 30 minutes, and the training items include gait training and hand coordination training; at the beginning of the rehabilitation training process, individual physiological signal data, such as heart rate, blood oxygen saturation and electromyography signals, are collected in real time; the collected individual physiological signal data are preprocessed, and a filtering algorithm is used to remove high-frequency noise and low-frequency drift, for example, a bandpass filter is used with a filtering frequency band of 0.5Hz to 40Hz; then the preprocessed individual physiological signal data are uniformly dimensionalized, and the minimum-maximum normalization method is used to map the data to the [0,1] interval; the processed individual physiological signal data are arranged in chronological order to generate a standardized physiological response data sequence.

[0075] S6. Use time series analysis and multimodal feature fusion methods to dynamically model the standardized physiological response data series and analyze abnormal states during rehabilitation training.

[0076] S6.1. Use time series analysis methods to dynamically model the standardized physiological response data series, extract the time-dependent characteristics of the physiological response, and obtain the physiological dynamic change characteristics.

[0077] Specifically, the time series analysis method is used to sort the standardized physiological response data sequence in ascending order according to the timestamp to ensure that the data time series is complete and continuous; a stationarity test is performed, such as using the ADF test method to determine whether the standardized physiological response data sequence is stationary. If it is not stationary, it is converted into a stationary series through the difference method; the appropriate time series model order is determined based on the autocorrelation coefficient diagram and the partial autocorrelation coefficient diagram of the standardized physiological response data sequence, such as determining that the autoregressive order in the autoregressive moving average model is 2 and the moving average order is 1; the least squares method or the maximum likelihood estimation method is used to estimate the parameters of the time series model to complete the time series model fitting; the fitting effect of the time series model is tested through residual sequence analysis to ensure that the residual sequence has no autocorrelation and the mean is close to zero; finally, based on the fitted time series model, the dependency characteristics of the physiological response data in the continuous time period are extracted, and physiological dynamic change characteristics reflecting the evolution of the individual physiological state are generated.

[0078] S6.2. Use the multimodal feature fusion method to fuse the physiological dynamic change characteristics with the rehabilitation response data sequence to generate a real-time physiological state feature vector.

[0079] Specifically, a multimodal feature fusion method is used to normalize the physiological dynamic change characteristics and the rehabilitation response data sequence respectively to ensure that each feature dimension has a unified dimension; the normalized physiological dynamic change characteristics and the rehabilitation response data sequence are aligned according to the time synchronization principle to form feature pairs at corresponding time points; a feature splicing method is used to merge the physiological dynamic change feature vectors and the rehabilitation response data sequence vectors at corresponding time points according to dimensions to generate a fused feature vector; the fused feature vector is subjected to dimensionality reduction processing, for example, the principal component analysis method is used to map the high-dimensional fusion features to a low-dimensional space to generate a real-time physiological state feature vector.

[0080] S6.3. Perform pattern recognition and trend assessment on real-time physiological state feature vectors, analyze abnormalities in state changes during rehabilitation training, and dynamically adjust the optimized individualized rehabilitation plan when abnormalities are detected.

[0081] Specifically, the support vector machine classification method is used to perform pattern recognition on the real-time physiological state feature vector to determine the category to which the current rehabilitation training state belongs; secondly, the trend indicators of the state feature vector, such as the moving average and slope, are calculated based on the sliding time window to evaluate the state change trend; the recognition result and the trend indicator are compared with the preset state change trend slope threshold to determine whether there is an abnormality; when an abnormality is detected, the rehabilitation control parameters in the optimized individualized rehabilitation plan are adjusted according to the type and degree of the abnormality, such as adjusting the training intensity or frequency; finally, the updated individualized rehabilitation plan is used for subsequent rehabilitation training.

[0082] It should also be explained that the specific steps of using the support vector machine classification method for pattern recognition to determine the category to which the current rehabilitation training status belongs are as follows: based on the rehabilitation response data sequence and the corresponding individual rehabilitation status assessment label, a training sample set is constructed, and each training sample contains a standardized real-time physiological state feature vector and a corresponding rehabilitation state category label; the feature values ​​of each dimension in the training sample set are normalized to ensure numerical stability during the support vector machine classification training process; the kernel function type used by the support vector machine classification method is selected, such as the radial basis kernel function, and the initial values ​​of the penalty coefficient and kernel parameters are set, such as the penalty coefficient is 1.0 and the radial basis function parameter is 0.1; the support vector machine classification method is trained using the training sample set to obtain the support vector, classification interval boundary and discriminant function parameters; after the training is completed, the standardized real-time physiological state feature vector is used as input and substituted into the trained support vector machine classification method to output the determined category label to which the current rehabilitation training status belongs; The specific steps of comparing the preset state change trend slope threshold to determine whether an abnormality exists include: calculating the change trend slope of the corresponding real-time physiological state feature vector within a continuous time window based on the standardized physiological response data sequence and the rehabilitation response data sequence, extracting the feature change value and time difference of each time period using an equally spaced sliding window method, and obtaining a trend indicator sequence; presetting the state change trend slope threshold based on the change of the corresponding physiological state characteristics in the rehabilitation response data sequence, and setting an abnormal upper limit slope threshold and an abnormal lower limit slope threshold, respectively, for example, the upper limit of the preset state change trend slope threshold is 0.15 and the lower limit is -0.15; combining the category label of the current rehabilitation training state output by the support vector machine classification method, calculating the normal change trend slope range when the category appears in the rehabilitation response data sequence; comparing the trend indicator value calculated within the current time window with the preset state change trend slope threshold one by one; if the trend indicator value exceeds the abnormal upper limit slope threshold or is lower than the abnormal lower limit slope threshold, it is determined that the current rehabilitation training state is abnormal; if all trend indicator values ​​are between the preset upper and lower limits of the state change trend slope threshold, it is determined that the current rehabilitation training state is normal.

[0083] This embodiment also provides a computer device, which is suitable for the rehabilitation nursing method for neurological patients based on multimodal data analysis, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the rehabilitation nursing method for neurological patients based on multimodal data analysis proposed in the above embodiment. The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0084] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the rehabilitation nursing method for neurological patients based on multimodal data analysis as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0085] In summary, the present invention achieves accurate identification of potential risks in the rehabilitation process by: deeply fusing the neuroplasticity index vector with the standardized unified multimodal feature representation, and combining it with statistical analysis methods to detect abnormal patterns; further generating personalized intervention suggestions and rehabilitation plans based on individual characteristics, effectively opening up the dynamic linkage mechanism between changes in neural function and behavioral feedback, breaking through the limitations of traditional reliance on a single modality and empirical evaluation, and realizing the transformation of the rehabilitation process from passive response to active prediction, significantly improving the scientific nature, timeliness and individual adaptability of the intervention, thereby accelerating neural function remodeling and reducing rehabilitation risks.

[0086] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A rehabilitation nursing method for neurological patients based on multimodal data analysis, characterized by: include, Collect multimodal data from neurology patients and perform standardized preprocessing to form a unified multimodal feature representation; Through the multimodal data fusion algorithm, neural function characteristics are extracted and calculated using weighted fusion and statistical analysis methods to obtain the neural plasticity index vector; Combining neuroplasticity index vectors with unified multimodal feature representations to perform multimodal abnormal pattern recognition analysis, obtain rehabilitation risk warning signals and personalized intervention recommendations, and generate individualized rehabilitation plans; Utilize the feedback loop mechanism to dynamically optimize and adaptively adjust the individualized rehabilitation plan to generate an optimized individualized rehabilitation plan; Real-time collection of physiological signals during individualized rehabilitation training based on optimized individualized rehabilitation programs, and pre-processing to generate standardized physiological response data sequences; Time series analysis and multimodal feature fusion methods are used to dynamically model standardized physiological response data sequences and analyze abnormal states during rehabilitation training.

2. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 1, characterized in that: The specific steps of forming a unified multimodal feature representation are as follows: Collect EEG, imaging, physiology, behavior and questionnaires from neurology patients to form multimodal data; Anomaly detection and interpolation repair methods are used to clean multimodal data and generate cleaned multimodal datasets; The cleaned multimodal dataset is format converted and time-series aligned to construct a structured time-synchronized multimodal data sample. Normalization and Z-score standardization methods are used to process it and generate standardized multimodal feature data. The feature mapping function is used to unify the dimensions and spatially project the standardized multimodal feature data to form a unified multimodal feature representation.

3. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 1, characterized in that: The specific steps of obtaining the neuroplasticity index vector are as follows: A feature screening algorithm is used to reduce the dimensionality of the unified multimodal feature representation, extract a subset of candidate features related to neural function, and screen out a set of highly relevant neural function features through correlation analysis and importance assessment. A weighted fusion method is used to fuse heterogeneous modal features in a set of highly correlated neural function features to construct a neural function fusion vector. Statistical analysis methods were used to perform attribution modeling and index calculation on the neural function fusion vector to obtain the neural plasticity index vector.

4. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 1, characterized in that: The specific steps for generating an individualized rehabilitation plan are as follows: The neuroplasticity index vector is fused with the unified multimodal feature representation to form a fused feature representation, and statistical analysis methods are used to detect abnormal patterns, identify abnormal features and generate rehabilitation risk warning signals; Based on the rehabilitation risk warning signals and combined with the individual characteristics of patients in the unified multimodal feature representation, personalized intervention recommendations are formulated and individualized rehabilitation plans are generated.

5. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 1, characterized in that: The specific steps for generating the optimized individualized rehabilitation plan are as follows: Collect real-time physiological response indicators during individualized rehabilitation training, use bandpass filtering to preprocess the signals, and construct a rehabilitation response data sequence; Using time series feature extraction algorithms and long short-term memory networks, dynamic feature extraction and trend modeling are performed on the rehabilitation response data sequence to generate individual rehabilitation status assessment results; Based on the differences between individual rehabilitation status assessment results and individual rehabilitation target parameters, the adaptive deviation of the individual rehabilitation program is analyzed, and an optimization algorithm is used to generate program adjustment suggestions; According to the program adjustment suggestions, the rehabilitation control parameters in the individualized rehabilitation program are updated, and the structure is optimized to generate an optimized individualized rehabilitation program.

6. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 5, characterized in that: According to the difference between the individual rehabilitation status assessment results and the individualized rehabilitation target parameters, the adaptive deviation of the individualized rehabilitation program is analyzed. The specific steps are as follows: Based on multimodal data, clinical information extraction and feature mapping methods are used to extract patients' diagnostic labels, functional status scores, and treatment plan elements, and construct individualized rehabilitation target parameters. Using structured extraction method, individualized rehabilitation goal parameters are structured and standardized rehabilitation goal vectors are generated. Based on the individual rehabilitation status assessment results, the rehabilitation response data sequence is analyzed using the time series regression modeling method to extract dynamic features and generate the current rehabilitation status feature vector; The current rehabilitation state feature vector is compared with the standardized rehabilitation target vector to obtain the index difference value, and the deviation analysis algorithm is used to quantify the degree of adaptive deviation between the individualized rehabilitation plan and the individualized rehabilitation target parameters.

7. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 1, characterized in that: Based on the optimized individualized rehabilitation plan, a personalized rehabilitation training plan is formulated and rehabilitation training is started. During the training process, individual physiological signal data is collected in real time, noise is removed and dimensions are unified to generate a standardized physiological response data sequence.

8. The rehabilitation nursing method for neurological patients based on multimodal data analysis according to claim 1, characterized in that: The method of using time series analysis and multimodal feature fusion to dynamically model the standardized physiological response data sequence and analyze abnormal states during rehabilitation training is as follows: Using time series analysis methods, the standardized physiological response data series is dynamically modeled, the time-dependent characteristics of the physiological response are extracted, and the physiological dynamic change characteristics are obtained; Using multimodal feature fusion methods, the physiological dynamic change characteristics are fused with the rehabilitation response data sequence to generate a real-time physiological state feature vector; Perform pattern recognition and trend evaluation on real-time physiological state feature vectors, analyze abnormalities in state changes during rehabilitation training, and dynamically adjust the optimized individualized rehabilitation plan when abnormalities are detected.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the neurology patient rehabilitation nursing method based on multimodal data analysis as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the neurology patient rehabilitation nursing method based on multimodal data analysis as described in any one of claims 1 to 8 are implemented.

Images