Dynamic monitoring system for spinal cord injury reconstruction process

Abstract

The invention relates to the field of medical dynamic monitoring, in particular to a dynamic monitoring system for a spinal cord injury reconstruction process. Comprising an acquisition and fusion module, a dynamic modeling module, a function evaluation module and an adaptation control module. According to the system, nerve and muscle electrophysiological signals are collected through a wearable surface electrode and a flexible microelectrode array, timing sequence errors are corrected in combination with a calibration sample, and multi-modal signal fusion is achieved. And constructing a topological sequence and time sequence dependence structure, and extracting an incubation period, an activation gradient and response dispersion by adopting a time sequence convolutional network. And identifying function transition and abnormal compensation behaviors through a function evaluation module, and outputting a neural pathway response sequence diagram. And the sampling frequency, the feature weight and the evaluation period are dynamically adjusted based on a Bayesian optimization strategy, and self-adaptive optimization of the monitoring strategy is realized. Therefore, precise modeling and efficient tracking of the neural pathway function reconstruction process are realized.

Description

Technical Field

[0001] The present invention relates to the field of medical dynamic monitoring, and in particular to a dynamic monitoring system for a spinal cord injury reconstruction process. Background Art

[0002] Functional reconstruction after spinal cord injury is a core challenge in the field of neurorehabilitation. Monitoring the recovery process of patients' neuromuscular function is of great significance for rehabilitation assessment and intervention decision-making.

[0003] Existing neuroelectrophysiological monitoring methods mostly rely on short-term intraoperative SEP / MEP measurements, which are limited by anesthesia status, operating experience, and stimulation parameters, making them difficult to use for long-term dynamic tracking after surgery. Although electromyographic acquisition has the advantage of being non-invasive, it has strong independence between channels and lacks expression of the response structure of the neuromuscular pathway, resulting in the inability to reconstruct the excitation-response path, and the evaluation results mostly remain at the signal amplitude level. At the same time, existing systems generally lack pathway structure modeling and adaptive evaluation mechanisms, making it difficult to support multi-cycle, cross-channel functional reconstruction analysis, and unable to identify functional transitions and compensatory behaviors, limiting their practicality in refined rehabilitation management.

[0004] To this end, the present invention proposes a dynamic monitoring system for the spinal cord injury reconstruction process. Summary of the Invention

[0005] The present invention provides a dynamic monitoring system for the spinal cord injury reconstruction process, aiming to achieve accurate evaluation and closed-loop feedback optimization of the neural pathway reconstruction process.

[0006] To achieve the above object, the present invention provides the following technical solutions: The present invention provides a dynamic monitoring system for a spinal cord injury reconstruction process, comprising: The acquisition and fusion module uses wearable surface electrodes and flexible microelectrode arrays to collect multi-channel neural and muscle electrophysiological signals. Based on a unified high-precision clock, it performs synchronous sampling, builds a compensation matrix based on calibration samples, and aligns and adjusts the signal timing structure to correct timing errors. This achieves registration and fusion of neural-muscle multimodal signals and outputs a unified timing matrix. The dynamic modeling module is used to rearrange the pathway structure of the unified timing matrix, construct a cross-node timing dependency sequence according to the neural excitation order, use a temporal convolutional network to extract features from each node time series, and output the latency, activation gradient, and response dispersion of each node; The functional assessment module constructs a time series trend based on the node response characteristics under multiple monitoring cycles, generates a neural pathway functional reconstruction score, identifies functional transitions and abnormal compensatory behaviors, and outputs a neural pathway response time series diagram; The adaptive control module dynamically adjusts the sampling frequency, feature weights, and evaluation period based on historical monitoring data and damage characteristics through the Bayesian decision update algorithm to achieve adaptive optimization of the monitoring strategy.

[0007] As a preferred technical solution of the invention, the calibration sample collection step includes: A functional electrical stimulation device is used to apply controlled current to the target muscle groups, inducing the limbs to perform passive standard movements; Collect electrophysiological signals and muscle tension change data during the induced movement; The collected data is used as a calibration sample to construct a compensation matrix to correct timing errors.

[0008] As a preferred technical solution of the invention, the step of correcting the timing error includes: The time difference between the neural activation time point and the corresponding electromyographic signal peak of each acquisition channel in the calibration sample is extracted to calculate the channel response delay value; Repeatedly sample the delay values ​​of each channel and take the average value to construct the channel delay compensation matrix; According to the channel delay, a time axis offset process is performed on the original sampling signal of each channel by using a linear translation method so as to align it to a unified response reference point; The phase difference between the neural activation time point and the peak of the electromyographic response is further extracted, and the sampling window position of the electromyographic signal is adjusted using a sliding average method; Delay compensation and phase correction are jointly applied to the original signal to achieve temporal alignment and individualized normalization of neuromuscular multimodal electrophysiological data.

[0009] As a preferred technical solution of the invention, the step of rearranging the path structure of the structurally unified timing matrix includes: According to the neuromuscular anatomical topological order corresponding to the electrode channels, the node channels in the timing matrix are sorted according to the neural excitation pathway; Identify the activation start time of each node in the response process and construct a multi-channel time series input with causal order as the modeling input of the temporal convolutional network.

[0010] As a preferred technical solution of the invention, the step of extracting features from the time series using a temporal convolutional network includes: Normalize and perform sliding window segmentation processing on the node time series signal; A one-dimensional temporal convolutional network is used to extract short-term dynamic features, and a residual structure is combined to preserve the signal change trend; Aggregation operations are performed on the output feature vectors to obtain the conduction latency, activation gradient and response dispersion indicators of each monitoring node.

[0011] As a preferred technical solution of the invention, the step of generating a neural pathway functional reconstruction score includes: Extract the response amplitude, latency and pathway activation sequence characteristics of each monitoring node under different monitoring cycles; Calculate the changing trend of each pathway indicator in the time dimension and compare it with historical samples; A reconstruction score is generated based on a statistical model to quantify the efficiency of neural pathway functional recovery.

[0012] As a preferred technical solution of the invention, the step of identifying functional transition includes: During continuous monitoring cycles, the evolution of response patterns at each node in the neural pathway is tracked; Detect the first occurrence of activation pathway shifts, significant increases in response synchrony, or significant decreases in latency; The above events are marked as triggering events of functional transition and are used to determine the key functional turning points in the rehabilitation process.

[0013] As a preferred technical solution of the invention, the step of identifying abnormal compensatory behavior includes: Analyze the emergence of unexpected activation pathways in neural pathways and monitor abnormal increases in activation intensity or asymmetric spread of node responses; Identify potential compensatory pathways based on temporal change maps and determine whether they deviate from normal activation logic in combination with prior structural models; This abnormal response pathway is labeled as a compensatory pathway to assist in the adjustment of rehabilitation strategies and risk warnings.

[0014] As a preferred technical solution of the invention, the adaptive optimization step includes: Collect historical monitoring data and patient injury characteristics to construct an objective function for optimizing parameter configuration; Based on the Bayesian optimization framework, variable modeling is performed on the sampling frequency, feature weights, and evaluation period control parameters; The expected improvement criterion is used to search for the optimal solution in the parameter space and update the current monitoring strategy to achieve closed-loop regulation and adaptive optimization.

[0015] The beneficial effects of the present invention are: 1. This invention achieves synchronous acquisition of multimodal nerve and muscle signals by integrating wearable surface electrodes with flexible microelectrode arrays. It also extracts inter-channel response delays and phase differences using calibration samples, constructs a channel compensation matrix and a sliding correction strategy, uniformly aligns the time axis and adjusts the phase of each channel signal, completes cross-channel error correction and standardization, and provides a highly consistent and personalized signal foundation for subsequent modeling. This significantly outperforms traditional independent channel acquisition and static compensation methods.

[0016] 2. Through the collaborative design of modeling modules and evaluation and control modules, this invention enables continuous modeling, dynamic evaluation, and real-time optimization of the functional status of neural pathways. This allows for timely identification of functional transitions and abnormal compensatory trends during monitoring, and for adjustments to sampling and evaluation strategies accordingly. Compared to traditional approaches that rely on single-cycle observations and static models, this invention accurately captures pathway change inflection points during multi-cycle recovery, improving responsiveness and adaptability to interventions during neural reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 It is a structural schematic diagram of a dynamic monitoring system for spinal cord injury reconstruction process according to the present invention; Figure 2 The present invention is a flow chart of a dynamic monitoring system for spinal cord injury reconstruction process. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0019] Example 1: Figure 1 As shown, the present invention provides a dynamic monitoring system for spinal cord injury reconstruction process, comprising: The acquisition and fusion module uses wearable surface electrodes and flexible microelectrode arrays to collect multi-channel neural and muscle electrophysiological signals. Based on a unified high-precision clock, it performs synchronous sampling, builds a compensation matrix based on calibration samples, and aligns and adjusts the signal timing structure to correct timing errors. This achieves registration and fusion of neural-muscle multimodal signals and outputs a unified timing matrix. Furthermore, the calibration sample collection step includes: A functional electrical stimulation device is used to apply controlled current to the target muscle groups, inducing the limbs to perform passive standard movements; Collect electrophysiological signals and muscle tension change data during the induced movement; The collected data is used as a calibration sample for correcting timing errors using an individualized compensation method.

[0020] Furthermore, the step of correcting the timing error includes: The time difference between the neural activation time point and the corresponding electromyographic signal peak of each acquisition channel in the calibration sample is extracted to calculate the channel response delay value; Repeatedly sample the delay values ​​of each channel and take the average value to construct the channel delay compensation matrix; According to the channel delay, a time axis offset process is performed on the original sampling signal of each channel by using a linear translation method so as to align it to a unified response reference point; The phase difference between the neural activation time point and the peak of the electromyographic response is further extracted, and the sampling window position of the electromyographic signal is adjusted using a sliding average method; Delay compensation and phase correction are jointly applied to the original signal to achieve temporal alignment and individualized normalization of neuromuscular multimodal electrophysiological data.

[0021] Specifically, wearable surface electromyography (EMG) electrodes are placed in the target muscle groups to collect surface electromyographic (EMG) signals generated by the muscles. Simultaneously, flexible microelectrode arrays are placed in key locations along the neural pathway to collect electrical signals generated by neural activity. These electrodes are connected to a synchronized acquisition device. The system aligns all acquisition channels based on a unified, high-precision clock, generating a cross-modal, multi-channel raw sampled data stream.

[0022] After initial use or re-positioning of electrodes, a functional electrical stimulation system applies a controlled current of a preset intensity to the target muscle group, inducing a passive standardized motor response. During the induced movement, electrophysiological signals and muscle tension change data are collected and used as calibration samples. The response delay m of the neuromuscular pathway is extracted based on the time difference between the neural signal activation moment and the peak myoelectric signal response after each electrical pulse in the calibration sample. Five repeated sampling groups are averaged for each channel to form a channel delay compensation matrix. The phase difference Δφ between the onset of electrical stimulation and the peak response is also recorded for subsequent compensation.

[0023] According to the delay difference of each channel in the delay matrix M, the linear translation compensation function is set: in is the original sampling signal, For the Channel delay, in ms.

[0024] After all channel signals are processed by this function, they are aligned to a unified response reference point on the time axis to achieve preliminary time alignment of cross-modal signals.

[0025] To correct the intention-response offset caused by nerve damage, the collected neural activation time points are recorded as , the peak point of the electromyographic signal is , the phase difference Δφ is used to correct the time window of the EMG signal and move the sampling segment forward or backward by sliding average to adjust the timing structure of the EMG response.

[0026] At the same time, the weight coefficient is calculated based on the mean value of the response amplitude of each channel , perform amplitude-weighted normalization on the signal amplitude and output the vector: All channel signals after time sequence correction and amplitude weighting are classified according to signal type. Assume that the number of neural signal channels is , the number of myoelectric signal channels is , the number of sampling points for each channel is , construct two matrices respectively: Neural Signal Matrix EMG signal matrix Based on the electrode layout and anatomical association, the two modal signals are aligned at the channel level. Each pair of channels forms a dual-channel timing block: All paired small matrices Splice in the row direction according to the anatomical order to construct a unified fusion matrix: By collaboratively collecting nerve and muscle electrophysiological signals through wearable surface electrodes and flexible microelectrode arrays, and combining high-precision clocks with personalized compensation strategies, precise alignment of different modal signals on the time axis is achieved, providing a structurally unified and temporally consistent data foundation for subsequent modeling and functional evaluation.

[0027] The dynamic modeling module is used to rearrange the pathway structure of the unified timing matrix, construct a cross-node timing dependency sequence according to the neural excitation order, use a temporal convolutional network to extract features from each node time series, and output the latency, activation gradient, and response dispersion of each node; Furthermore, the step of constructing the node timing dependency sequence includes: The pathway structure of the unified timing matrix is ​​rearranged, and the node channels are sorted according to the order of the neural excitation path; Extract the activation start time in each node response and the delay relationship between nodes; Construct an input sequence with topological constraints and temporal dependencies for modeling by the temporal convolutional network.

[0028] Specifically, the fusion matrix output by the acquisition and fusion module , where each row corresponds to the time series electrophysiological signal of an electrode channel. For each monitoring node, directly extract its complete time series data, that is, The node signal is expressed as , without any dimensionality reduction or truncation processing, ensuring the integrity and continuity of the time series signal.

[0029] Based on the anatomical structure of the nervous system and the neural excitation pathways, the channels corresponding to different modal signals are rearranged according to their physiological locations and conduction order. For example, the nerve channel signals located in the upper spinal cord are arranged first, followed by the corresponding distal electromyographic channel signals. The rearranged signal sequence forms a two-dimensional matrix with the number of rows corresponding to the number of nodes and the number of columns corresponding to the number of time points: in Indicates the The node index after sorting, .

[0030] For each rearranged channel , extract its activation starting time point The activation time point is determined by the following method: First, the time series of each node is segmented by sliding window, and the average energy of each segment is calculated. , the energy calculation formula is: in is the window length, The sliding start time point.

[0031] Define global energy threshold , activation time point For the first one to make time point.

[0032] Then, calculate the timing delay between adjacent channels: Finally, the timing matrix after the path is rearranged The corresponding node activation time point and delayed sequence Together as input, we construct a structured input sequence for temporal convolutional network modeling.

[0033] While maintaining the original multi-channel time series information, this input structure embeds the topological constraints of the neural excitation path and the temporal dependencies between nodes, which helps subsequent models to accurately extract features such as pathway latency, activation gradient, and response discreteness.

[0034] Furthermore, the step of extracting features from the time series using a temporal convolutional network includes: Normalize and perform sliding window segmentation processing on the node time series signal; A one-dimensional temporal convolutional network is used to extract short-term dynamic features, and a residual structure is combined to preserve the signal change trend; Aggregation operations are performed on the output feature vectors to obtain the conduction latency, activation gradient and response dispersion indicators of each monitoring node.

[0035] Specifically, according to the rearranged timing matrix , combined with the activation start time of each node Activation delay sequence between adjacent nodes , construct a three-channel input tensor , the channel data are as follows: Channel 1: Normalize the original time series signal of each node and map it to the [0,1] interval as the main input; Second channel: Each node is marked as "1" within a certain time window after the activation start time, and marked as "0" at other time points, forming a binary mask sequence to represent the significant response segment; Channel 3: For each node, fill in its activation delay value relative to the upstream node , this value serves as a reference quantity for topological-temporal dependence and is replicated in the time dimension to form a constant value sequence.

[0036] The input tensor is sliced ​​along the time dimension using a fixed window length, generating multiple overlapping time series segments. Each segment retains its original multi-channel structure, and all segments are combined into a batch input to meet the input requirements of the time series convolutional network. To ensure input data stability, the time series signal is normalized before sliding the window, outliers are trimmed using threshold detection, and any missing segments are padded with zeros.

[0037] The processed time series is used as input to a one-dimensional temporal convolutional network (TCN), which extracts the time-frequency dynamic features of the signal within each time window while preserving the overall trend of the signal. The TCN outputs a feature vector corresponding to each sliding window, reflecting the local activation pattern and dynamic changes.

[0038] For multiple sliding window feature vectors of the same monitoring node, the sliding windows are arranged in sequence to form a feature matrix ,in is the number of sliding windows, is the dimension of a single sliding window feature vector. Based on the aggregated features, the following quantitative indicators are calculated: Conduction latency: The feature dimension that reflects the activation intensity sets the threshold ; Find the first sliding window in time sequence The characteristic value is greater than The index of the conduction latency is the time point corresponding to ; Activation gradient: Select the feature dimension that reflects the activation intensity and calculate the first-order difference on the time series ,The activation gradient is defined as the average value of the difference sequence, which reflects the overall activation rate change of the node during the monitoring period; Response discreteness: the order of values ​​of the same feature dimension on the time axis The standard deviation is calculated as a response dispersion indicator, which represents the stability of the node's activation state throughout the cycle.

[0039] The adaptability of signal preprocessing was improved through normalization and sliding window processing. The short-term dynamic features were extracted using a one-dimensional time series convolutional network, and the overall trend was retained in combination with the residual structure. Ultimately, the key response characteristics of each monitoring node were accurately quantified, providing a reliable data basis for subsequent functional evaluation.

[0040] The functional assessment module constructs a time series trend based on the node response characteristics under multiple monitoring cycles, generates a neural pathway functional reconstruction score, identifies functional transitions and abnormal compensatory behaviors, and outputs a neural pathway response time series diagram; Furthermore, the step of generating a neural pathway functional reconstruction score includes: Extract the response amplitude, latency and pathway activation sequence characteristics of each monitoring node under different monitoring cycles; Calculate the changing trend of each pathway indicator in the time dimension and compare it with historical samples; A reconstruction score is generated based on a statistical model to quantify the efficiency of neural pathway functional recovery.

[0041] Specifically, within each preset monitoring cycle, the output results of the aforementioned dynamic modeling module are called to extract the following three types of characteristic indicators of all monitoring nodes: conduction latency, activation gradient, and response discreteness.

[0042] The three types of indicators of each monitoring node in the current cycle are organized into vectors 、 、 , respectively, representing the Conduction latency, activation gradient, and response dispersion under 1 cycle.

[0043] During multiple consecutive rehabilitation monitoring cycles, the evolution trend of each type of indicator on the time axis is constructed separately: right The latency value of each monitoring node in the dataset is used for linear regression fitting to obtain the slope of the latency change of the node; right For each element of The speed of improvement within a cycle; right Calculate the variation range and mean fluctuation amplitude to determine whether the response stability has improved.

[0044] In order to obtain quantitative results of the degree of functional recovery of the overall neural pathway, the target change direction of each type of indicator is set: shortening of latency, strengthening of gradient, and decreasing of dispersion, and the expected improvement range is defined.

[0045] The trend values ​​of each node on the three indicators are standardized to The interval is weighted and summed according to the preset weights to calculate the overall neural pathway reconstruction score .

[0046] Rating value The changing trend of is used to evaluate the patient's rehabilitation progress in each cycle. A higher value indicates that the neural pathway activation is closer to the healthy pathway state, which serves as the basis for adjusting the rehabilitation plan and evaluating the staged efficacy.

[0047] Furthermore, the step of identifying functional transition includes: During continuous monitoring cycles, the evolution of response patterns at each node in the neural pathway is tracked; Detect the first occurrence of activation pathway shifts, significant increases in response synchrony, or significant decreases in latency; The above events are marked as triggering events of functional transition and are used to determine the key functional turning points in the rehabilitation process.

[0048] During the set continuous monitoring period, the three indicators of conduction latency, activation gradient and response dispersion of each node are extracted to form a periodic sequence. ; Conduction latency , construct a time series for each monitoring node, and use the first-order difference method to calculate its rate of change. If A sudden drop occurs, and the drop exceeds the set threshold , it is determined to be a latent period jump; Compare the activation path sequence: in the cycle and , construct the corresponding path and ,If its directed connection relationship changes, that is, a path jump occurs or a secondary path is activated in advance, it is marked as an ,activation path transfer event; In addition, the Pearson correlation coefficient matrix between the responses of each periodic node is calculated. If the average synchronization is significantly improved, in this embodiment, if the overall correlation coefficient is improved by more than 0.3, it is determined to be a response synchronization transition; A cycle that meets any of the conditions is identified as a functional transition trigger cycle, and the corresponding node and time point will be marked as a key inflection point in the response timing diagram.

[0049] Furthermore, the step of identifying abnormal compensatory behavior includes: Analyze the emergence of unexpected activation pathways in neural pathways and monitor abnormal increases in activation intensity or asymmetric spread of node responses; Identify potential compensatory pathways based on temporal change maps and determine whether they deviate from normal activation logic in combination with prior structural models; This abnormal response pathway is labeled as a compensatory pathway to assist in the adjustment of rehabilitation strategies and risk warnings.

[0050] For each cycle, analyze the set of newly activated nodes outside the neural activation path , compare it with the predefined normal path model, if If there is no structural connection between the intermediate node and the main path, or its activation delay is abnormally shortened (e.g., less than 5ms), it is marked as an unexpected activation path; Activation gradient The node values ​​were Z-score normalized. If the activation gradient value of a single node in a pathway exceeded 2 standard deviations from the mean, it was considered to be an abnormally elevated activation intensity. If multiple nodes are activated in a clustered manner on one side in space, the response dispersion will be Asymmetric response diffusion is identified when one side significantly expands while the other side does not respond or responds attenuatedly. The above-mentioned abnormal activation pathways are constructed as independent pathways, and based on the structural model, it is determined whether they deviate from the logic of the neuromuscular physiological pathway. If so, the pathway is marked as an abnormal compensatory pathway and used to indicate potential risks.

[0051] Finally, the main activation pathways within each cycle were plotted with the time period as the horizontal axis and the anatomical location of the node as the vertical axis; For the identified functional transition events, mark the “transition point” on the corresponding cycle node; The identified abnormal compensatory pathways are represented by dotted lines or special colors, indicating their first appearance cycle and pathway structure; An interactive neural pathway response timing diagram is formed to provide clinicians with a basis for judging rehabilitation turning points and serve as a visual reference tool for adjusting rehabilitation intervention strategies.

[0052] The adaptive control module dynamically adjusts the sampling frequency, feature weights, and evaluation period based on historical monitoring data and damage characteristics through the Bayesian decision update algorithm to achieve adaptive optimization of the monitoring strategy.

[0053] Furthermore, the adaptive optimization step includes: Collect historical monitoring data and patient injury characteristics to construct an objective function for optimizing parameter configuration; Based on the Bayesian optimization framework, variable modeling is performed on the sampling frequency, feature weights, and evaluation period control parameters; The expected improvement criterion is used to search for the optimal solution in the parameter space and update the current monitoring strategy to achieve adaptive optimization of the monitoring strategy.

[0054] The system first constructs an objective function to evaluate the pros and cons of the current monitoring strategy based on historical monitoring data and patient injury characteristics. Among them, the input parameters Corresponding to sampling frequency , feature weight and evaluation cycle The output result is a comprehensive score value, which represents the overall monitoring effect under the current parameter combination. The scoring function is as follows: : The reconstruction score of this cycle output by the neural pathway function assessment module; : The width of the confidence interval of the continuous evaluation results reflects the stability of the indicator; : The estimated resource consumption of the system under the current parameter combination (acquisition duration, number of electrodes used, etc.); : The weight coefficient calculated based on the normalization of the partial derivative of the scoring function is as follows: This setting ensures that the scoring function can reflect the impact of different factors on the effectiveness of rehabilitation monitoring strategies in real time.

[0055] The system is based on an existing historical data sample set , using Gaussian process regression for the objective function Modeling is performed in the parameter space to obtain the mean and variance of each parameter combination to estimate its potential monitoring effect.

[0056] Using the expected improvement criterion as a sampling strategy, we search for the next set of optimal parameters in the parameter space that can maximize the improvement of monitoring effect. .

[0057] After the optimal solution is determined, the system automatically updates the sampling frequency, feature weighting coefficient and evaluation execution cycle of the next monitoring cycle.

[0058] After each monitoring cycle, the system adds the current parameter combination and evaluation results to the sample set for the next round of Bayesian model updates, forming a closed-loop mechanism for continuous iterative optimization. To prevent frequent policy changes, a parameter change threshold is set, meaning that the policy is only updated when the improvement exceeds the set threshold.

[0059] Example 2: In a patient with spinal cord injury who was admitted to a rehabilitation center, a dynamic monitoring system for the reconstruction process of spinal cord injury according to the present invention was applied. Figure 2 , the following is the actual deployment and usage process: First, flexible microelectrode arrays were placed on both sides of the patient's lumbar spine to collect real-time neural electrical signals from the peripheral nerve pathways in the spinal cord injury area. Simultaneously, wearable surface electrodes were attached to key motor muscles, including the bilateral rectus femoris, iliopsoas, and tibialis anterior, to collect myoelectric signals. All acquisition channels were connected to a unified high-precision acquisition interface, and the system automatically synchronized the signals across the channels.

[0060] Considering that the patient is unable to actively complete standard movements, the clinical rehabilitation therapist uses electrical stimulation equipment to apply controllable current to the areas innervated by the patient's femoral nerve and sciatic nerve, inducing passive movements such as knee extension and ankle dorsiflexion, and synchronously collects nerve signals, electromyographic signals, and changes in muscle tension during the movement. The system records the activation delay time, signal strength, and response morphology under multiple rounds of electrical stimulation, constructs an individual channel response delay parameter and signal phase difference model, and automatically generates a corresponding compensation matrix. Subsequently, the system performs delay adjustment, phase correction, and channel weight normalization on the original signal, outputting a set of nerve-muscle fusion signal matrices with unified structure and time alignment.

[0061] After data fusion is complete, the system rearranges all channel signals according to the conduction sequence of "spinal nerves - nerve roots - proximal muscle groups - distal muscle groups" based on the patient's anatomical information, constructing a structural topological sequence of nerve excitation and muscle response. The system also scans each channel signal, identifying the starting time of the first continuous activation waveform, and calculates the response delay between channels to generate a time-dependent diagram of the excitation path.

[0062] The system inputs the rearranged signal matrix and activation delay structure into a pre-trained temporal convolutional network model, extracting characteristic indicators such as conduction latency, activation trend change rate, and response stability for each node. After the model output is generated, the results of the current cycle are compared with the patient's data from the previous cycle. If a significant shortening of the latency, a shift in the activation sequence, or the emergence of a new signal channel activation pathway is detected, the system will flag a possible "functional transition" in that cycle. If abnormal discharges of certain muscle groups are detected without stimulation from higher-level nerves, this is recorded as a risk of "abnormal compensation" and a doctor will be prompted to review the situation.

[0063] At the same time, the system automatically retrieves the past four monitoring data, and combines it with label data such as the patient's initial post-injury grade, surgical method, and recovery speed, and uses the Bayesian optimization strategy to adaptively adjust the monitoring cycle, feature selection priority, and sampling channel. When the system enters the next round of monitoring, it can focus on more representative nodes and features, thereby improving evaluation efficiency and accuracy.

[0064] Finally, the patient experienced two typical activation path transition events at 3 and 6 weeks after surgery, respectively. The activation latency of the rectus femoris and tibialis anterior muscles was shortened by more than 45%, the synchronization of the neural pathways was enhanced, and the functional score curve output by the system gradually increased. Based on this, the doctor decided to carry out passive walking assistance training for the lower limbs in advance. After the system prompted the existence of an abnormal compensatory path in the 8th week, the doctor promptly fine-tuned the rehabilitation strategy to avoid the risk of muscle overload during training.

[0065] This second embodiment fully demonstrates the application effect of the system of the present invention in the entire process from data acquisition, standardized fusion, dynamic modeling, function identification to monitoring strategy optimization in a real clinical environment, proving its practicality and innovation in spinal cord injury reconstruction and rehabilitation assessment.

[0066] Finally, it should be noted that the above descriptions are merely preferred embodiments 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 aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A dynamic monitoring system for spinal cord injury reconstruction process, characterized in that: include: The acquisition and fusion module uses wearable surface electrodes and flexible microelectrode arrays to collect multi-channel neural and muscle electrophysiological signals. Based on a unified high-precision clock, it performs synchronous sampling, builds a compensation matrix based on calibration samples, and aligns and adjusts the signal timing structure to correct timing errors. This achieves registration and fusion of neural-muscle multimodal signals and outputs a unified timing matrix. The dynamic modeling module is used to rearrange the pathway structure of the unified timing matrix, construct a cross-node timing dependency sequence according to the neural excitation order, use a temporal convolutional network to extract features from each node time series, and output the latency, activation gradient, and response dispersion of each node; The functional assessment module constructs a time series trend based on the node response characteristics under multiple monitoring cycles, generates a neural pathway functional reconstruction score, identifies functional transitions and abnormal compensatory behaviors, and outputs a neural pathway response time series diagram; The adaptive control module dynamically adjusts the sampling frequency, feature weights, and evaluation period based on historical monitoring data and damage characteristics through the Bayesian decision update algorithm to achieve adaptive optimization of the monitoring strategy.

2. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The calibration sample collection step includes: A functional electrical stimulation device is used to apply controlled current to the target muscle groups, inducing the limbs to perform passive standard movements; Collect electrophysiological signals and muscle tension change data during the induced movement; The collected data is used as a calibration sample to construct a compensation matrix to correct timing errors.

3. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The step of correcting the timing error includes: The time difference between the neural activation time point and the corresponding electromyographic signal peak of each acquisition channel in the calibration sample is extracted to calculate the channel response delay value; Repeatedly sample the delay values ​​of each channel and take the average value to construct the channel delay compensation matrix; According to the channel delay, a time axis offset process is performed on the original sampling signal of each channel by using a linear translation method so as to align it to a unified response reference point; The phase difference between the neural activation time point and the peak of the electromyographic response is further extracted, and the sampling window position of the electromyographic signal is adjusted using a sliding average method; Delay compensation and phase correction are jointly applied to the original signal to achieve temporal alignment and individualized normalization of neuromuscular multimodal electrophysiological data.

4. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The step of rearranging the path structure of the structurally unified timing matrix comprises: According to the neuromuscular anatomical topological order corresponding to the electrode channels, the node channels in the timing matrix are sorted according to the neural excitation pathway; Identify the activation start time of each node in the response process and construct a multi-channel time series input with causal order as the modeling input of the temporal convolutional network.

5. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The step of extracting features from the time series using a temporal convolutional network includes: Normalize and perform sliding window segmentation processing on the node time series signal; A one-dimensional temporal convolutional network is used to extract short-term dynamic features, and a residual structure is combined to preserve the signal change trend; Aggregation operations are performed on the output feature vectors to obtain the conduction latency, activation gradient and response dispersion indicators of each monitoring node.

6. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The step of generating a neural pathway functional reconstruction score comprises: Extract the response amplitude, latency and pathway activation sequence characteristics of each monitoring node under different monitoring cycles; Calculate the changing trend of each pathway indicator in the time dimension and compare it with historical samples; A reconstruction score is generated based on a statistical model to quantify the efficiency of neural pathway functional recovery.

7. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The step of identifying functional transition comprises: During continuous monitoring cycles, the evolution of response patterns at each node in the neural pathway is tracked; Detect the first occurrence of activation pathway shifts, significant increases in response synchrony, or significant decreases in latency; The above events are marked as triggering events of functional transition and are used to determine the key functional turning points in the rehabilitation process.

8. The dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The step of identifying abnormal compensatory behavior includes: Analyze the emergence of unexpected activation pathways in neural pathways and monitor abnormal increases in activation intensity or asymmetric spread of node responses; Identify potential compensatory pathways based on temporal change maps and determine whether they deviate from normal activation logic in combination with prior structural models; This abnormal response pathway is labeled as a compensatory pathway to assist in the adjustment of rehabilitation strategies and risk warnings.

9. A dynamic monitoring system for spinal cord injury reconstruction process according to claim 1, characterized in that: The adaptive optimization step comprises: Collect historical monitoring data and patient injury characteristics to construct an objective function for optimizing parameter configuration; Based on the Bayesian optimization framework, variable modeling is performed on the sampling frequency, feature weights, and evaluation period control parameters; The expected improvement criterion is used to search for the optimal solution in the parameter space and update the current monitoring strategy to achieve closed-loop regulation and adaptive optimization.

Images