Rehabilitation training system for motion function reconstruction of spinal cord injury paraplegia patient

Abstract

The embodiment of the invention discloses a rehabilitation training system for motor function reconstruction of a spinal cord injury paraplegia patient. A specific implementation mode of the system comprises a bedside lower limb passive training system, a lower limb function reconstruction and intensive training system, a server and a doctor mobile terminal, wherein the server is configured to control the bedside lower limb passive training equipment to assist in performing at least one of the following training at a preset frequency: passive joint training, mind-driven auxiliary exercise training and standing-bed training; the bedside lower limb passive training equipment is configured to execute the following steps: generating training stage conversion test information; the server is further configured to generate second training control information for the lower limb function reconstruction and intensive training system to perform at least one motion function reconstruction training under closed-loop spinal cord stimulation gait adjustment. According to the embodiment, the risk of secondary injury caused in the motion function reconstruction training process is improved.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a rehabilitation training system for the reconstruction of motor function in patients with spinal cord injury and paraplegia. Background Technology

[0002] With the continuous advancement of rehabilitation medicine and neuroengineering technology, rehabilitation training systems have been widely applied in the functional recovery treatment of spinal cord injury patients. Rehabilitation training systems for the reconstruction of motor function in spinal cord injury paraplegic patients provide users with a full-cycle rehabilitation service, from early bedside passive activities to later active gait reconstruction, generating training control instructions corresponding to the patient's spinal cord injury status. Currently, the common approach to implementing spinal cord injury rehabilitation training is to divide the rehabilitation process into independent stages. First, doctors develop a fixed bedside passive training plan based on experience (such as using a CPM machine for joint movement). After the patient has been hospitalized for a period of time or the doctor's manual assessment deems the conditions suitable, the training is manually switched to lower limb functional reconstruction training (such as exoskeleton robot gait training or electrical stimulation training).

[0003] However, when implementing rehabilitation training using the above methods, the following technical problems often arise: The rehabilitation process is fragmented into independent stages, with distinct barriers between them. Because the transition from bedside passive training to functional reconstruction training relies primarily on periodic manual assessments by physicians, lacking real-time objective data support, the switching of training modes often lags behind the actual recovery progress of the patient's neurological function. Switching too early may lead to motor injury or the solidification of abnormal gait due to insufficient residual muscle strength; switching too late misses the optimal window for neuroplasticity, resulting in low rehabilitation efficiency and prolonged recovery time. Furthermore, existing functional reconstruction training systems (such as spinal cord stimulation devices) often employ open-loop control, outputting electrical pulses at fixed frequencies and parameters, failing to dynamically adjust stimulation based on real-time changes in the patient's functional state during training, thus increasing the risk of secondary injury.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not form prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a rehabilitation training system and a method for sending training phase transition test information for the reconstruction of motor function in patients with spinal cord injury and paraplegia, in order to address one or more of the technical intentions mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a rehabilitation training system for the reconstruction of motor function in paraplegic patients with spinal cord injury. The system includes: a bedside lower limb passive training system, a lower limb function reconstruction and strengthening training system, a server, and a doctor's mobile terminal. The doctor's mobile terminal is configured to send spinal cord injury information corresponding to a user identifier to the server. The server is configured to perform the following steps: based on the spinal cord injury information, generate first training control information corresponding to the bedside lower limb passive training system; send the first training control information to the bedside lower limb passive training system corresponding to the user identifier to control the bedside lower limb passive training device in the system to assist in at least one of the following training methods at a preset frequency: passive joint training, mind-driven assisted movement. The system includes both dynamic training and standing training. The bedside lower limb passive training device in the aforementioned bedside lower limb passive training system is configured to perform the following steps: detecting training phase transitions during thought-driven assisted movement training to generate training phase transition test information; sending the training phase transition test information to a doctor's mobile terminal, so that the doctor's mobile terminal can send the user identifier and the training phase transition test information to the server; the server is further configured to generate second training control information based on the training phase transition test information, and send the second training control information to the lower limb function reconstruction and strengthening training system corresponding to the user identifier, so that the lower limb function reconstruction and strengthening training system can perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation based on the second training control information.

[0008] Secondly, some embodiments of this disclosure provide a method for sending training phase transition test information. The method includes: performing training phase transition detection in mind-driven assisted movement training to generate training phase transition test information; and sending the training phase transition test information to a doctor's mobile terminal so that the doctor's mobile terminal can send the user identifier and the training phase transition test information to the server.

[0009] The above-described embodiments of this disclosure have the following beneficial effects: The rehabilitation training system for motor function reconstruction in spinal cord injury paraplegic patients according to some embodiments of this disclosure reduces motor injury or gait fixation during motor rehabilitation training, shortens the patient's recovery period, and reduces the risk of secondary injury during training. Specifically, the reasons for motor injury or gait fixation, prolonged recovery period, and increased risk of secondary injury during motor rehabilitation training are: the rehabilitation process is fragmented into independent stages, with gaps between different stages. Since the transition from bedside passive training to functional reconstruction training mainly relies on periodic manual assessments by doctors, lacking real-time objective data support, the switching of training modes often lags behind the actual recovery process of the patient's neurological function. If the switch is too early, insufficient residual muscle strength may lead to motor injury or gait fixation; if the switch is too late, the optimal window of neuroplasticity is missed, resulting in low rehabilitation efficiency and prolonged recovery period. Moreover, existing functional reconstruction training systems (such as spinal cord electrical stimulation devices) often adopt an open-loop control mode, that is, outputting electrical pulses according to fixed frequency and parameters, which cannot dynamically adjust stimulation based on real-time changes in the patient's functional state during training, increasing the risk of secondary injury. Based on this, some embodiments of the present disclosure provide a rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia, comprising: a bedside lower limb passive training system, a lower limb function reconstruction and strengthening training system, a server, and a doctor's mobile terminal, wherein: the doctor's mobile terminal is configured to send spinal cord injury information corresponding to a user identifier to the server. Then, the server is configured to perform the following steps: based on the spinal cord injury information, generate first training control information corresponding to the bedside lower limb passive training system. Thus, personalized early first training control information can be generated according to the patient's spinal cord injury information, i.e., the patient's specific injury condition, ensuring safety and targeted intervention in the early stages of rehabilitation. Subsequently, the first training control information is sent to the bedside lower limb passive training system corresponding to the user identifier to control the bedside lower limb passive training device in the system to assist in at least one of the following training methods at a preset frequency: passive joint training, mind-driven assisted movement training, and standing training. Therefore, the bedside lower limb passive training device can be driven at a preset frequency to assist in at least one of the following training methods: passive joint training, mind-driven assisted movement training, and standing training, with at least one targeted training method based on the patient's injury condition. Subsequently, the bedside lower limb passive training device in the aforementioned bedside lower limb passive training system is configured to perform the following steps: First, during mind-driven assisted movement training, a training phase transition detection is performed to generate training phase transition test information. Thus, training phase transition test information can be generated during each mind-driven assisted movement training session to ensure that the patient's potential to enter the next stage can be identified as early as possible.The second step involves sending the aforementioned training phase transition test information to the doctor's mobile terminal, which then forwards the user identifier and training phase transition test information to the server. This allows the server to generate second training control information for entering the next stage of training, namely lower limb function reconstruction and strengthening training. This ensures timely intervention within the optimal window of neural plasticity, reducing the risk of motor injury or gait stagnation due to premature or delayed switching, and prolonging the patient's recovery period. The server is then further configured to generate the second training control information based on the training phase transition test information and send it to the lower limb function reconstruction and strengthening training system corresponding to the user identifier. This allows the system to perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation based on the second training control information. Thus, the lower limb function reconstruction and strengthening training system can perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation, dynamically adjusting stimulation based on real-time changes in the patient's functional state during training, reducing the risk of secondary injury. Because real-time training phase transition detection was performed during each mind-driven assisted movement training session using the bedside lower limb passive training system, training phase transition test information reflecting the patient's true neurological recovery status was generated. Based on this training phase transition test information, the server dynamically generated secondary training control information and controlled the lower limb function reconstruction and strengthening training system to perform closed-loop spinal cord stimulation gait regulation. This eliminated the lag in rehabilitation phase transitions, ensuring timely intervention during the optimal window of neuroplasticity and reducing problems such as motor injury or abnormal gait solidification caused by premature or late switching, thus prolonging the patient's recovery period. Furthermore, by performing at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation and sensing changes in status, the risk of secondary injury during motor function reconstruction training was reduced. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0011] Figure 1 This is an architecture diagram of an exemplary system for a rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia, based on the present disclosure. Figure 2 Flowcharts of some embodiments of the training phase conversion test information sending method according to this disclosure. Detailed Implementation

[0012] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0013] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0014] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0015] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0016] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0017] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Figure 1 An exemplary system architecture 100 for a rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia, to which some embodiments of the present disclosure may be applied, is shown.

[0019] like Figure 1As shown, the system architecture 100 may include: a bedside lower limb passive training system 103, a lower limb function reconstruction and strengthening training system 104, a server 102, and a doctor's mobile terminal 101. The server is connected to the bedside lower limb passive training system 103, the lower limb function reconstruction and strengthening training system 104, and the doctor's mobile terminal 101 via a network connection. The network connection method may include various connection types, such as wired, wireless communication links, or fiber optic cables. The bedside lower limb passive training system 103 may be a system that includes bedside lower limb passive training equipment (such as a bedside lower limb exoskeleton robot), an external decoder, a surface electromyography (EMG) electrode array, and an optional electroencephalogram (EEG) acquisition module (if thought-driven, it includes a brain-computer interface). The surface EMG electrode array can collect surface EMG signals generated by the patient in real time when attempting limb movement. The surface EMG electrode array may consist of multiple flexible dry or wet electrodes, attached according to the anatomical locations of the major muscle groups of the lower limb (such as the quadriceps, hamstrings, tibialis anterior, and gastrocnemius). The aforementioned external decoder may include a signal amplification and filtering module, an analog-to-digital converter (ADC), and an embedded microprocessor (MCU / FPGA), and a decoder with a pre-trained neural intent recognition algorithm model (such as Agile Temporal Convolutional Neural Network (ATCN)). The aforementioned lower limb function reconstruction and strengthening training system 104 may be a system including a weight-reducing gait training device, a gait rehabilitation robot, and an actively trained lower limb exoskeleton robot. The aforementioned weight-reducing gait training device may be a rehabilitation weight-reducing gait trainer. The aforementioned gait rehabilitation robot may be a device to assist patients with gait disorders in rehabilitation training. The aforementioned actively trained lower limb exoskeleton robot may be an actively assisted lower limb rehabilitation exoskeleton robot that senses the patient's movement intent in real time and provides assistance according to the intent. The aforementioned doctor's mobile terminal 101 may be a terminal device used by the doctor (e.g., a terminal computing device, such as a computer or mobile phone). The rehabilitation training system for the reconstruction of motor function in patients with spinal cord injury and paraplegia also includes a home training device management system 105, which includes wearable lower limb assistive movement devices and a user terminal.

[0020] In some embodiments, the aforementioned doctor's mobile terminal can be configured to send spinal cord injury information corresponding to a user identifier to the aforementioned server. This spinal cord injury information can represent the degree of spinal cord injury suffered by the patient. Specifically, it can be a muscle strength rating; for example, in the acute / early stage of spinal cord injury, the information can be represented as muscle strength grade 0 or 1. In the hospital rehabilitation stage (mid-stage), the information can be represented as muscle strength grade 2 or 3. In the home rehabilitation stage, the information can represent muscle strength grade 3 or higher. The user identifier can be a patient's identifier, such as a patient's mobile phone number.

[0021] In some embodiments, the server is configured to perform the following steps: The first step is to generate first training control information corresponding to the bedside lower limb passive training system based on the aforementioned spinal cord injury information. In practice, the server 102 can query the training control information corresponding to the muscle strength grade represented by the spinal cord injury information in the preset training control knowledge base as the first training control information. The preset training control knowledge base can be a pre-defined knowledge base containing various correspondences between muscle strength grades and training control information. The training control information can be a set of digital instructions sent to the bedside lower limb passive training system to execute personalized exercise intervention strategies. For example, the first training control information corresponding to muscle strength grade 1 can be a set of digital instructions representing the personalized exercise intervention strategy as follows: "9:00 AM: Passive joint movement training (exoskeleton execution) 20 minutes. 10:30 AM: Attempt mind-driven training (brain-computer interface + exoskeleton) 3 sets, 5 attempts per set. 3:00 PM: Standing training 15 minutes (combined with spinal cord electrical stimulation)."

[0022] The second step involves sending the aforementioned first training control information to the bedside lower limb passive training system corresponding to the aforementioned user identifier. This controls the bedside lower limb passive training device within the system to assist in at least one of the following training methods at a preset frequency: passive joint training, thought-driven assisted movement training, and standing training. The preset frequency can be once daily or once weekly. Passive joint training involves the device forcibly moving the patient's lower limb joints in a full-range or limited-range periodic movement according to preset kinematic parameters, without the patient needing (or being unable to) actively exert force. Thought-driven assisted movement training involves detecting the patient's motor intention (via EEG signals) or muscle activation signals generated by residual neural pathways (via sEMG signals). The device only provides the corresponding assistive torque to help the patient complete the movement when a clear motor intention is confirmed. In other words, "the machine assists when the patient wants to move; the machine waits when the patient does not move." Standing training involves gradually raising the patient from a supine position to an upright position (or near-upright position).

[0023] In some embodiments, the above-described bedside lower limb passive training system can be configured to perform the following steps: The first step is to perform training phase transition detection in mind-driven assisted movement training to generate training phase transition test information.

[0024] In some optional implementations of certain embodiments, the above-described bedside lower limb passive training device can be further configured to perform training phase transition detection during mind-driven assisted movement training through the following steps to generate training phase transition test information: Step one involves receiving the EEG decoding confidence level from an external decoder and collecting leg electromyography (EMG) signals via a surface EMG electrode array during each thought-driven assisted movement training session. The EEG decoding confidence level and leg EMG signals are then stored in an EEG decoding confidence level queue and a leg EMG signal queue, respectively. In practice, during each thought-driven assisted movement training session, the external decoder sends the latest EEG decoding confidence level to the bedside lower limb passive training device at a fixed frequency (e.g., 50Hz or 100Hz). This EEG decoding confidence level allows the external decoder's built-in intention recognition model (e.g., Agile Temporal Convolutional Neural Network (ATCN)) to perform real-time inference on the EEG features within a sliding window, outputting the probability value that the patient currently has a specific movement intention (e.g., "knee extension"). The leg EMG signals are electrical signals collected from the skin surface caused by the bioelectrical activity generated by motor neurons when the lower limb muscles contract or attempt to contract.

[0025] Step 2: Based on the EEG decoding confidence queue and the leg EMG signal queue, generate training device conversion prompts and send these prompts to the doctor's mobile terminal. In practice, the ratio of the number of EEG decoding confidence scores greater than a preset confidence score in the EEG decoding confidence queue to the total number of EEG decoding confidence scores in the queue can be determined as the target ratio. Then, in response to determining that the target ratio is greater than the preset ratio, for each EEG decoding confidence score greater than the preset confidence score in the EEG decoding confidence queue, perform the following steps: First, determine the sequence number of the EEG decoding confidence score in the EEG decoding confidence queue as the target sequence number (e.g., 5). Then, determine the sequence number range (e.g., 5-7) following the target sequence number, including the time window (e.g., time window 3). Then, determine at least one leg EMG signal corresponding to the target sequence number range in the leg EMG signal queue as at least one detected leg EMG signal. Subsequently, in response to determining that at least one detected leg EMG signal exceeds a preset EMG activation threshold, the EEG decoding confidence level can be determined as the effective EEG decoding confidence level. Finally, the ratio of the number of effective EEG decoding confidence levels to the total number of EEG decoding confidence levels in the EEG decoding confidence level cohort can be determined as the reference ratio. In response to determining that the reference ratio is greater than a preset value, the text information representing the transition from bedside passive training to functional reconstruction training can be determined as the training device transition prompt information.

[0026] Step 3: In response to the detection of a training device conversion prompt, the device works in conjunction with the suspension weight reduction device to perform the test task. The suspension weight reduction device can be a suspension rehabilitation training device.

[0027] Step four, in executing the test task, perform the following test steps: The first sub-step involves acquiring brainwave signals generated by the discharge of nerve cell groups in the user's motor cortex and supplementary motor area using intravascular EEG electrodes to obtain a potential amplitude sequence. The aforementioned intravascular EEG electrodes can be endovascular EEG acquisition electrodes. The potential amplitude sequence can be a set of potential amplitudes acquired according to a set of sampling timestamps over a continuous time period during the execution of the test task.

[0028] The second sub-step involves acquiring an electromyographic (EMG) signal sequence using a surface electromyography (SEMG) electrode array. This EMG signal sequence can be a set of EMG signals acquired over a continuous time period during the test task, using the same set of sampling timestamps as the acquired potential amplitude sequence. The EMG signal amplitude represents the voltage value (e.g., 0.2 millivolts (mV)) of the electrophysiological signal generated by muscle activity at a given moment. For example, an EMG signal sequence could be "(0.2 (mV), -0.4 (mV), 0.5 (mV), -0.6 (mV), 0.3 (mV))". The EMG signal is an alternating current signal, typically containing alternating positive and negative voltage values. Positive and negative values ​​represent the direction of muscle contraction (e.g., a positive value represents the opposite direction of a contraction, and a negative value represents the other pole of the contraction).

[0029] The third sub-step involves acquiring inertial data corresponding to each moving part through an inertial measurement unit (IMU). Each piece of inertial data can include acceleration, angular velocity, and attitude angle.

[0030] The fourth sub-step involves acquiring pressure information sets for the left and right feet using a pressure sensor array. Each potential amplitude in the aforementioned potential amplitude sequence corresponds to a sampling time, and each electromyographic signal in the aforementioned electromyographic signal sequence corresponds to a sampling time.

[0031] Step 5: Based on the above potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group and right foot pressure information group, generate training phase transition test information.

[0032] In some optional implementations of certain embodiments, the above-mentioned bedside lower limb passive training device can be further configured to generate training phase transition test information based on the above-mentioned potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group, and right foot pressure information group through the following steps: Step one involves inputting the potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group, and right foot pressure information group into the modality-specific feature extraction layer of a pre-trained rehabilitation assessment model. This yields potential feature information, electromyographic feature information, kinematic feature information, and dynamic feature information. The rehabilitation assessment model includes the modality-specific feature extraction layer, feature fusion and interaction layer, and functional index regression layer. The modality-specific feature extraction layer comprises the potential feature extraction sub-network, electromyographic feature extraction sub-network, kinematic feature extraction sub-network, and pressure feature extraction sub-network. The potential feature extraction sub-network can be a convolutional neural network that takes the potential amplitude sequence as input and outputs feature vectors representing the frequency and amplitude characteristics of the potential signal (i.e., potential feature information). The electromyographic feature extraction sub-network can be a temporal convolutional network that takes the electromyographic signal sequence as input and outputs feature vectors representing electromyographic features (such as the intensity, frequency, and time-series characteristics of electromyographic activity) (i.e., electromyographic feature information). The aforementioned kinematic feature extraction subnetwork can be a convolutional neural network that takes various inertial data as input and outputs feature vectors representing kinematic features (such as posture changes, direction of movement, angle changes, etc.). Similarly, the aforementioned pressure feature extraction subnetwork can be a recursive neural network that takes left foot pressure information sets and right foot pressure information sets as input and outputs feature vectors representing pressure features (such as the distribution of foot pressure during gait, differences in load between the left and right feet, etc.).

[0033] Step two involves feeding the aforementioned potential feature information, electromyographic feature information, kinematic feature information, and dynamic feature information into the feature fusion and interaction layer to obtain cross-modal joint behavioral feature information. The feature fusion and interaction layer can be a fusion layer that integrates the potential feature information, electromyographic feature information, kinematic feature information, and dynamic feature information (e.g., weighted averaging, splicing, etc.) to obtain the cross-modal joint behavioral feature information. The feature fusion and interaction layer can be a fusion layer for performing feature fusion.

[0034] Step 3: Input the aforementioned cross-modal joint behavioral feature information into the aforementioned functional index regression layer to obtain the comprehensive walking ability level, neuromotor control score, weight-bearing propulsion score, and gait stability score. The aforementioned functional index regression layer can be a fully connected layer that takes the cross-modal joint behavioral feature information as input and outputs the comprehensive walking ability level, neuromotor control score, weight-bearing propulsion score, and gait stability score. The aforementioned comprehensive walking ability level can be a grading score of the patient's overall independent walking ability (e.g., level 0: unable to walk; level 1: requires assistance to walk independently). The aforementioned neuromotor control score can be a score assessing the central nervous system's control over the lower limb muscles (e.g., the higher the score, the stronger the control over the lower limb muscles). The aforementioned weight-bearing propulsion score represents the patient's weight-bearing capacity (e.g., the higher the score, the stronger the weight-bearing capacity). The aforementioned gait stability score can assess the patient's ability to maintain balance, resist disturbances, and prevent falls during walking.

[0035] Step four: The above-mentioned comprehensive walking ability level, the above-mentioned neuromotor control score, the above-mentioned weight-bearing propulsion score, and the above-mentioned gait stability score are determined as the training phase transition test information.

[0036] In some optional implementations of certain embodiments, the above-mentioned bedside lower limb passive training device can be further configured to input the potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group, and right foot pressure information group into the modality-specific feature extraction layer of a pre-trained rehabilitation assessment model through the following steps to obtain potential feature information, electromyographic feature information, kinematic feature information, and dynamic feature information: Step 1: Input the above potential amplitude sequence into the potential feature extraction subnetwork included in the modality-specific feature extraction layer to obtain potential feature information. The modality-specific feature extraction layer includes the above potential feature extraction subnetwork, electromyographic feature extraction subnetwork, kinematic feature extraction subnetwork and pressure feature extraction subnetwork.

[0037] Step 2: Input the above electromyographic signal sequence into the above electromyographic feature extraction subnetwork to obtain electromyographic feature information.

[0038] Step 3: Input the above inertial data into the kinematic feature extraction subnetwork to obtain kinematic feature information.

[0039] Step four: Input the left foot pressure information group and the right foot pressure information group into the pressure feature extraction sub-network mentioned above to obtain dynamic feature information.

[0040] The second step is to send the aforementioned training phase transition test information to the doctor's mobile terminal, so that the doctor's mobile terminal can send the aforementioned user identifier and the aforementioned training phase transition test information to the aforementioned server.

[0041] In some embodiments, the server is further configured to generate second training control information based on training phase transition test information, and to send the second training control information to the lower limb function reconstruction and strengthening training system corresponding to the user identifier, so that the lower limb function reconstruction and strengthening training system can perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation based on the second training control information.

[0042] In some optional implementations of certain embodiments, the server described above can generate second training control information based on the transformed test information during the training phase through the following steps: The first step, in response to determining that the overall walking ability level is greater than or equal to the first preset level and less than or equal to the second preset level, is to perform the following steps: The first step involves determining that the neuromotor control score included in the training phase transition test information is less than the preset neuromotor control score, and then defining the preset parameter control information corresponding to the active training lower limb exoskeleton robot as the second training control sub-information. This preset parameter control information corresponding to the active training lower limb exoskeleton robot can be the configuration information for configuring the active training lower limb exoskeleton robot to assist the patient in intention-driven walking training. For example, the preset parameter control information corresponding to the actively trained lower limb exoskeleton robot can be: "Training Mode": "Intention-Driven On-Demand Assist (AAN)", "Trigger Settings": {"EMG Activation Threshold": "12μV", / / The robot recognizes even the slightest muscle exertion from the patient, "Response Delay": "<40ms" / / Ensures that the robot's movements are synchronized with the patient's intentions}, "Assistance Strategy": {"Torque Gain Coefficient": "0.65", / / The robot automatically supplements the 65% of force missing from the patient, "Trajectory Stiffness": "Low" / / Allows for slight deviations in the patient's movements, without forced locking}, "Safety Limits": {"Maximum Output Torque": "40Nm" / / Prevents excessive assist force from straining the joints}.

[0043] The second step involves determining that the weight-bearing propulsion score included in the training phase transition test information is less than the preset weight-bearing propulsion score, and then defining the preset parameter control information corresponding to the weight-loss walking training device as the second training control sub-information. This preset parameter control information corresponding to the weight-loss walking training device can be configuration information for configuring the weight-loss walking training device to assist the patient in weight-loss supportive walking training.

[0044] The third step involves determining that the gait stability score included in the training phase transition test information is less than the preset gait stability score, and then defining the preset parameter control information of the gait rehabilitation robot as the second training control sub-information. This preset parameter control information of the gait rehabilitation robot can be the configuration information for configuring the gait rehabilitation robot to assist the patient in machine-assisted gait training.

[0045] The second step is to determine at least one of the identified second training control sub-information as the second training control information.

[0046] The intended application scenario is rehabilitation training for patients with spinal cord injury and paraplegia, particularly focusing on gait adjustment and motor function reconstruction in scenarios such as weight-bearing gait training, machine-assisted gait training, and intention-driven gait training. An accompanying technical challenge is that when using weight-bearing gait devices, rehabilitation robots, or exoskeleton robots to assist patients in gait training, traditional open-loop electrical stimulation cannot dynamically adjust based on real-time gait status due to significant individual patient differences, diverse gait patterns, and the prevalence of abnormal gait (such as poor stability, insufficient muscle strength, and clonus). This results in poor stimulation effects and difficulty in effectively improving gait coordination and motor control. The following characteristics are required for this application scenario: Spinal cord injury rehabilitation training needs to perceive the patient's gait status and motor intention in real time and dynamically adjust spinal cord electrical stimulation parameters based on abnormal gait events (such as gait instability, insufficient hip flexion, and clonus) to improve gait coordination, enhance motor control, and prevent abnormal patterns, thereby achieving personalized, closed-loop rehabilitation training.

[0047] In some optional implementations of certain embodiments, the aforementioned lower limb function reconstruction and strengthening training system is further configured to perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait modulation based on second training control information through the following steps: The first step involves using at least one of the following devices—a weight-bearing gait training device, a gait rehabilitation robot, or an active training lower limb exoskeleton robot—based on second training control information to assist spinal cord injury paraplegic patients in performing at least one of the following motor function reconstruction training methods: weight-bearing supportive gait training, machine-assisted gait training, and intention-driven gait training. Specifically, the weight-bearing supportive gait training can involve partially or completely offsetting the patient's weight through a suspension system (sling), allowing the lower limb joints (especially the hip, knee, and ankle) to perform gait training under low load. The machine-assisted gait training can involve using a rehabilitation robot (usually end-drive or exoskeleton-type) to passively or semi-passively drive the patient's lower limbs to walk according to a preset standard physiological gait trajectory. The intention-driven gait training can be an active assisted lower limb rehabilitation training that senses the patient's movement intention in real time and provides assistance based on that intention; that is, training based on the patient's active movement intention (via electromyography (sEMG) signals or human-machine interaction force detection), providing on-demand assistance only when the patient's effort is insufficient.

[0048] In the second step, during each of the weight-bearing supported walking training, machine-assisted gait training, and intention-driven walking training, the aforementioned closed-loop spinal cord stimulator is configured to perform the following closed-loop spinal cord stimulation gait modulation process: The first sub-step involves receiving plantar pressure distribution information and joint angle monitoring data from the weight-loss gait training device during weight-loss supportive gait training. Based on this information, abnormal gait detection is performed to obtain abnormal gait detection information. The plantar pressure distribution information can represent the distribution of plantar pressure. The joint angle monitoring data can be the instantaneous rotation angle values ​​of key angle points (such as the ankle joint) in the coronal plane corresponding to a gait cycle. In practice, a human stability determination method based on the center of plantar pressure (CoP) can be used to determine the plantar pressure distribution information and generate a first stable state label (e.g., "stable" or "unstable"). Then, the difference between the maximum and minimum instantaneous rotation angle values ​​is determined as the target difference. In response to determining that the target difference is greater than a preset difference, the unstable label is determined as the second stable state label. In response to determining that an unstable label exists between the first and second stable state labels, information representing an abnormal gait event is determined as abnormal gait detection information.

[0049] The second sub-step involves characterizing abnormal gait events based on abnormal gait detection information and increasing the frequency of spinal cord stimulation to improve gait coordination.

[0050] The third sub-step involves receiving knee joint torque, knee joint torque amplitude, and knee joint torque fluctuation frequency from the gait rehabilitation robot during machine-assisted gait training. Based on the knee joint torque, knee joint torque amplitude, and knee joint torque fluctuation frequency, the spinal cord electrical stimulation mode is adjusted to a preset electrical stimulation mode, wherein the preset electrical stimulation mode is one of the following: hip flexion enhancement mode and clonic seizure blocking mode. In practice, in response to the determination that the knee joint torque is less than a first preset threshold (e.g., 10 Nm), the knee joint torque amplitude is less than a first preset amplitude (e.g., 5 Nm), and the knee joint torque fluctuation frequency is less than a first preset frequency (e.g., 1 Hz), the spinal cord stimulation mode is adjusted to the hip flexion enhancement mode (i.e., adjusted to improve the strength and control of hip flexion for "insufficient activity," helping patients better complete gait or other movements; the parameters for the hip flexion enhancement mode can be "stimulation mode: hip flexion enhancement mode, parameter preset: 100ms short burst, 100 Hz, pulse width 200 μs, electrode configuration: L2-L3 bipolar"). In response to the determination that the knee joint torque is greater than a second preset threshold (e.g., 15 Nm), the knee joint torque amplitude is less than a second preset amplitude (e.g., 15 Nm), and the knee joint torque fluctuation frequency is less than a second preset frequency (e.g., 2 Hz), the spinal cord stimulation mode is adjusted to the clonic seizure blocking mode. (That is, for "clonic seizures", the electrical stimulation mode is adjusted to the clonic seizure blocking mode to reduce or prevent abnormal clonic seizures and improve the stability and safety of movement. The parameters corresponding to the clonic seizure blocking mode can be "stimulation mode: high frequency blocking mode, parameter preset: 120Hz continuous, pulse width 120μs, amplitude increase preset amplitude, electrode configuration: S1-S2 wide field").

[0051] The fourth sub-step is to respond to receiving exoskeleton assistance demand information sent by the active training type lower limb exoskeleton robot during the intention-driven walking training process and to determine that the exoskeleton assistance demand information represents an active knee extension intention, and to stimulate the femoral nerve at a preset frequency.

[0052] The above-mentioned technical solution and related content, as an inventive point of this disclosure, solve the technical problem of "poor stimulation effect, making it difficult to effectively improve gait coordination and motor control ability". Factors leading to poor stimulation effect and difficulty in effectively improving gait coordination and motor control ability are often as follows: When using weight-reducing walking devices, rehabilitation robots, or exoskeleton robots to assist patients in gait training, due to large individual differences among patients, diverse gait patterns, and the tendency for abnormal gait (such as poor stability, insufficient muscle strength, clonus, etc.), traditional open-loop electrical stimulation cannot dynamically adjust according to real-time gait status, resulting in poor stimulation effect and difficulty in effectively improving gait coordination and motor control ability. If the above factors are solved, the stimulation effect can be improved, thereby improving gait coordination and motor control ability. To achieve this effect, firstly, based on second training control information, at least one of the following devices—weight-reducing walking training device, gait rehabilitation robot, and active training lower limb exoskeleton robot—is used to assist spinal cord injury paraplegic patients in performing at least one of the following motor function reconstruction training: weight-reducing supportive walking training, machine-assisted gait training, and intention-driven walking training. Then, during each of the weight-bearing supported gait training, machine-assisted gait training, and intention-driven gait training, the aforementioned closed-loop spinal cord stimulator is configured to perform the following closed-loop spinal cord stimulation gait modulation process: First, in response to receiving plantar pressure distribution information and joint angle monitoring data from the weight-bearing gait training device during the weight-bearing supported gait training, abnormal gait detection is performed based on the plantar pressure distribution information and joint angle monitoring data to obtain abnormal gait detection information. Second, based on the abnormal gait detection information indicating the presence of gait abnormalities, the spinal cord stimulation frequency is increased to improve gait coordination. Thus, abnormal gait can be detected using plantar pressure distribution and joint angle data, and the spinal cord stimulation frequency can be adjusted according to the detection results to improve the stimulation effect and gait coordination. The third step involves receiving knee joint torque, knee joint torque amplitude, and knee joint torque fluctuation frequency from the gait rehabilitation robot during machine-assisted gait training. Based on these parameters, the spinal cord stimulation mode is adjusted to a preset mode, which is one of the following: hip flexion enhancement mode or clonic seizure blocking mode. This allows for adjustment of the spinal cord stimulation mode (e.g., switching to hip flexion enhancement mode or clonic seizure blocking mode) to address specific gait issues and improve knee joint motor control and stability. The fourth step involves receiving exoskeleton assistance demand information from the active training lower limb exoskeleton robot during intention-driven gait training and determining that this information represents an intention to actively extend the knee. The femoral nerve is then stimulated at a preset frequency.Therefore, spinal cord stimulation can be driven by the intention to actively extend the knee, ensuring that the adjustment of spinal cord stimulation is based on the patient's current movement needs and intentions. Through intention-driven stimulation, the patient's gait needs can be better matched, thereby improving gait coordination and enhancing the patient's motor control ability.

[0053] In some optional implementations of some embodiments, the rehabilitation training system for reconstructing motor function in patients with spinal cord injury and paraplegia described above further includes a home training equipment management system 105, which includes wearable lower limb assistive exercise devices and a user terminal. The aforementioned doctor's mobile terminal is configured to send home training request information corresponding to a user identifier to the aforementioned server. This home training request information includes a user identifier, user address information, muscle strength level information, and a doctor's signature. The user address information can represent the user's residential address. The muscle strength level information can represent a muscle strength grade. The doctor's signature can be an electronic signature. The aforementioned user terminal can be a terminal device used by the patient. The aforementioned home training device management system can be a system that includes wearable lower limb assistive exercise devices and a user terminal.

[0054] The server is further configured to perform the following steps: The first step is to verify the doctor's signature in the received home training request message and obtain the verification result. In practice, a preset public key can be used to decrypt the doctor's signature to obtain the decrypted hash value.

[0055] The second step is to determine that the decrypted hash value is stored in a preset database and to identify the information indicating that the verification has been passed as the verification result.

[0056] The third step, in response to the confirmation that the verification result indicates successful verification and that the level represented by the aforementioned muscle strength level information is greater than the pre-approval level, involves querying the home-use wearable lower limb assistive exercise device repository corresponding to the aforementioned user address information. In practice, Geographic Information System (GIS) technology or simple string matching can be used to find the home-use wearable lower limb assistive exercise device repository corresponding to the user address information. This repository can be a storage facility for home-use wearable lower limb assistive exercise devices. These devices can be community / home-based assistive lower limb exoskeleton robots used for assisted training (such as lightweight exoskeleton-assisted walking, upper limb function training, and daily living activities training).

[0057] The fourth step is to dispatch one of the wearable lower limb assistive devices from the warehouse to the location corresponding to the user's address information. In practice, the server can control a dispatching device (e.g., an unmanned logistics vehicle) to dispatch one of the wearable lower limb assistive devices from the warehouse to the location corresponding to the user's address information.

[0058] The fifth step involves receiving an activation request from the user terminal corresponding to the home-wearable lower limb assistive exercise device, sending preset initial training control information to the device, and completing the configuration to control the device to perform assisted training. The preset initial training control information can be preset configuration information for configuring the home-wearable lower limb assistive exercise device to perform assisted training.

[0059] The aforementioned wearable lower limb assistive exercise device is configured to perform continuous fatigue monitoring and feedback processing during assisted training: The first step involves real-time collection of motion data and electromyography (EMG) signals using a sensor array installed on a wearable lower limb assistive device at home. The collected motion data and EMG signals are then stored in preset motion data queues and EMG signal queues, respectively. The sensor array includes an accelerometer and an EMG sensor. The motion data can be acceleration in three dimensions (X, Y, and Z directions).

[0060] The second step involves retrieving motion data sequences and electromyographic (EMG) signal sequences from a preset motion data queue and an EMG signal queue at preset time intervals. In practice, a preset number of motion data points can be retrieved from the preset motion data queue from beginning to end and arranged according to their order within the queue to obtain the motion data sequence. Similarly, a preset number of EMG signals can be retrieved from the preset EMG signal sequence from beginning to end and arranged according to their order within the preset EMG signal queue to obtain the EMG signal sequence.

[0061] The third step involves fatigue detection processing based on the motion data sequence and electromyography (EMG) signal sequence to generate and broadcast fatigue alerts. In practice, for every two consecutive motion data points in a pre-defined motion data queue, the rate of change of acceleration between these two consecutive data points can be defined as the acceleration rate of change. Then, the obtained acceleration rates of change are arranged to obtain an acceleration rate of change sequence. Next, the variance of the acceleration rate of change sequence can be calculated. The aforementioned acceleration rate of change can be the magnitude of the difference between the acceleration vectors included in the two consecutive motion data points. As an example, the two consecutive motion data points can be respectively... , . It can represent the acceleration in the X direction included in the first of two consecutive motion data. It can represent the acceleration in the Y direction included in the previous motion data. It can represent the acceleration in the Z direction included in the first of two consecutive motion data. It can represent the acceleration in the X direction included in the latter of two consecutive motion data. It can represent the acceleration in the Y direction included in the subsequent motion data. It can represent the acceleration in the Z direction included in the latter of two consecutive motion data. The rate of change of acceleration can be expressed as... express:

[0062] Next, the aforementioned executing entity can use the root mean square (RMS) of the electromyographic (EMG) signal amplitudes included in the EMG signal sequence. Then, in response to determining that the variance of the acceleration change is greater than a first preset threshold and the RMS is less than a second preset threshold, text information representing a fatigue state is determined as fatigue warning information. In response to determining that the variance of the acceleration change is greater than the first preset threshold and the RMS is greater than the second preset threshold, text information representing an early stage of fatigue is determined as fatigue warning information.

[0063] Home-based rehabilitation training for patients with lower limb motor dysfunction often presents the following technical challenges: Patients transitioning to home-based rehabilitation need to obtain wearable lower limb assistive devices from a doctor's prescription and transport them to their homes themselves, resulting in a lack of on-site equipment delivery services and a poor user experience. Furthermore, self-operation of the rehabilitation devices by patients often lacks guidance from professional therapists, and because patients cannot accurately perceive their own muscle fatigue levels, overtraining can easily lead to secondary injuries. Therefore, this application scenario requires the following characteristics: remote rehabilitation management needs to achieve fully automated integration from doctor's prescription to equipment dispatch, ensuring that those with adequate muscle strength can obtain suitable equipment in a timely manner. The remote home training process needs to monitor physiological data in real time and provide intelligent alerts to ensure the safety and effectiveness of patient training in unsupervised environments.

[0064] The above-described technical solution and its related content, as an inventive point of this disclosure, solve the technical problem of "low user experience and easy overtraining leading to secondary injury." Factors contributing to a low user experience and easy overtraining leading to secondary injury are often as follows: When patients switch to home rehabilitation training, they need to obtain home-wearable lower limb assistive exercise devices from a doctor's prescription and transport them to their homes themselves, lacking on-site equipment service, resulting in a low user experience. Furthermore, training by operating the rehabilitation devices independently often lacks guidance from professional rehabilitation therapists, and because patients cannot accurately perceive their own muscle fatigue levels, overtraining leading to secondary injury is likely. Solving these factors can improve the user experience and reduce the risk of secondary injury due to overtraining. To achieve this effect, the system also includes a home training device management system, which includes home-wearable lower limb assistive exercise devices and a user terminal: the doctor's mobile terminal is configured to send home training request information corresponding to a user identifier to the server, wherein the home training request information includes a user identifier, user address information, muscle strength level information, and doctor's signature. The server is further configured to perform the following steps: verify the doctor's signature in the received home training request information and obtain a verification result. In response to the verification result indicating successful verification and that the level represented by the muscle strength level information is greater than the pre-approval level, query the home-wearable lower limb assistive exercise device warehouse corresponding to the user's address information. Schedule one home-wearable lower limb assistive exercise device from the warehouse to the location corresponding to the user's address information. Thus, the system achieves fully automated connection from doctor's prescription to device scheduling, providing on-site device service and improving user experience. Then, in response to receiving the activation request information corresponding to the home-wearable lower limb assistive exercise device sent by the user, send preset initial training control information to the home-wearable lower limb assistive exercise device and complete the configuration to control the home-wearable lower limb assistive exercise device to perform assisted training. Next, the aforementioned home-wearable lower limb assistive exercise device is configured to perform continuous fatigue monitoring and feedback processing during assisted training: a sensor array installed on the device collects motion data and electromyography (EMG) signals in real time, storing these data in preset motion data queues and EMG signal queues, respectively. This allows for continuous real-time acquisition of motion data and EMG signals during assisted training, capturing the actual muscle responses and limb movements. Then, at preset time intervals, motion data sequences and EMG signal sequences are retrieved from the preset motion data queues and EMG signal queues. Finally, fatigue detection processing is performed based on the motion data sequences and EMG signal sequences to generate and broadcast fatigue alerts.Therefore, fatigue alerts can be generated and broadcast based on motion data sequences and electromyographic signal sequences. The system can issue timely voice warnings before the patient realizes they are exhausted or that improper movement could lead to injury. Because it employs continuous monitoring and feedback processing to broadcast fatigue alerts during assisted training, it replaces the on-site monitoring of a rehabilitation therapist. This allows for real-time monitoring and intervention of fatigue risks during home training, while also reducing the incidence of secondary injuries due to overtraining.

[0065] For the intended application scenario: rehabilitation or physical training using wearable lower limb assistive devices at home often presents the following technical challenges: fatigue detection methods based on fixed, universal thresholds cannot adapt to individual physiological differences among users (such as age and weight), leading to a disconnect between fatigue assessment standards and actual capabilities. Simultaneously, motion artifacts caused by the ease with which wearable devices can loosen in a home environment can severely contaminate electromyographic signals, and single-modal data struggles to distinguish between "true muscle fatigue" and "motor compensation," resulting in frequent false fatigue alarms. If the threshold is set too high, forced training under extreme fatigue can easily lead to muscle strains or secondary joint injuries, increasing the risk of secondary damage. If the threshold is too low, frequent false fatigue alarms occur. Scenario characteristics: Individual physiological differences exist among users (such as age and weight), resulting in varying fatigue standards. Dynamically adjusting the threshold allows for assessment of whether a state of fatigue has been entered based on each individual's actual physical condition and their own fatigue standards.

[0066] In some optional implementations of certain embodiments, the wearable lower limb assistive exercise device in the aforementioned home training device management system is further configured to perform continuous fatigue monitoring and feedback processing during the assisted training process through the following steps: The first step, during assisted training, involves acquiring a fatigue threshold assessment profile corresponding to the user's identifier, and generating a dynamic fatigue threshold based on this profile. The fatigue threshold assessment profile can be a structured dataset describing the patient's basic information and physical condition (e.g., age, weight, medical history, exercise habits). This fatigue threshold assessment profile includes at least one of the following: age, weight, and upper limit of exercise duration. In practice, the fatigue threshold assessment profile can be input into a preset fatigue threshold calculation formula to obtain the dynamic fatigue threshold. As an example, the preset fatigue threshold calculation formula can be:

[0067] The above The above The above Preset weight values ​​can be used. The upper limit of the exercise duration mentioned above can be expressed in duration (e.g., hours). The dynamic fatigue threshold mentioned above can be used to characterize the degree of fatigue when a user enters a state of fatigue.

[0068] The second step involves real-time acquisition of motion data and electromyographic signals, and storing the acquired motion data and electromyographic signals into preset motion data queues and electromyographic signal queues, respectively.

[0069] The third step involves retrieving motion data sequences and electromyographic signal sequences from preset motion data queues and electromyographic signal queues at preset time intervals.

[0070] The fourth step involves inputting the aforementioned electromyographic (EMG) signal sequence into the artifact removal EMG feature extraction module of a pre-trained fatigue monitoring model to obtain EMG artifact removal feature extraction information. The fatigue monitoring model includes an artifact removal EMG feature extraction module, a motion temporal feature extraction module, a spatiotemporal cross-attention fusion module, and an adaptive fatigue state detection module. The artifact removal EMG feature extraction module comprises convolutional layers and a Transformer encoder. In practice, the time-frequency features of the EMG signal sequence can be extracted using convolutional layers, and the Transformer encoder (with its self-attention mechanism, which automatically focuses on the most critical parts of the signal (i.e., important muscle activation information) while filtering out irrelevant motion artifacts) can be used to encode these features. The final result is a high-dimensional feature vector representing the muscle activation pattern after filtering out irrelevant motion artifacts, which serves as the EMG artifact removal feature extraction information.

[0071] The fifth step involves inputting the aforementioned motion data sequence into the motion temporal feature extraction module to obtain kinematic temporal feature information. This module can be a Long Short-Term Memory (LSTM) network or a GRU model that takes the motion data sequence as input and outputs kinematic temporal feature information. The kinematic temporal feature information can be a high-dimensional vector representing kinematic features (e.g., smoothness features during the action).

[0072] Step 6: The aforementioned EMG artifact removal feature extraction information and kinematic temporal feature information are input into the aforementioned spatiotemporal cross-attention fusion module. Feature fusion is performed through the spatiotemporal cross-attention mechanism to obtain motion state coupled feature information. The aforementioned spatiotemporal cross-attention fusion module can be a fusion layer that fuses the EMG artifact removal feature extraction information and kinematic temporal feature information through cross-attention calculation (the cross-attention mechanism allows one modality (e.g., kinematic data) to "interrogate" another modality (e.g., EMG data), thereby obtaining a more meaningful feature representation). The aforementioned motion state coupled feature information can be a feature vector obtained by fusing the EMG artifact removal feature extraction information and kinematic temporal feature information.

[0073] Step 7: Input the aforementioned motion state coupling feature information and dynamic fatigue threshold into the adaptive fatigue state detection module to obtain fatigue detection information. The adaptive fatigue state detection module may include a regression layer. In practice, the adaptive fatigue state detection module takes the motion state coupling feature information (i.e., the fused feature vector) as input and a fatigue level score as output (e.g., a score from 0 to 10). Next, the adaptive fatigue state detection module compares the fatigue level score with the dynamic fatigue threshold. If the fatigue level score is greater than or equal to the dynamic fatigue threshold, a preset fatigue identifier is identified as fatigue detection information. If the fatigue level score is less than the dynamic fatigue threshold, a preset non-fatigue identifier is identified as fatigue detection information.

[0074] Step 8: Based on the fatigue detection information mentioned above, generate and broadcast fatigue alert information. In practice, in response to determining that the fatigue detection information is a preset fatigue identifier, the text information representing the fatigue state is determined as the fatigue alert information.

[0075] The above-mentioned technical solution and its related content, as an inventive point of this disclosure, solve the technical problem of "increased risk of secondary injury and frequent fatigue false alarms." Factors leading to increased risk of secondary injury and frequent fatigue false alarms are often as follows: fatigue detection methods based on fixed universal thresholds cannot adapt to the physiological differences between individual users (such as age, weight, etc.), resulting in a disconnect between fatigue judgment standards and actual ability; simultaneously, motion artifacts caused by wearable devices easily loosening in home environments can severely contaminate electromyographic signals, and single-modal data makes it difficult to distinguish between "true muscle fatigue" and "motor compensation," leading to frequent fatigue false alarms. If the threshold is set too high, forced training under extreme fatigue can easily cause muscle strain or secondary joint injury, increasing the risk of secondary injury. If the threshold is too low, it leads to frequent fatigue false alarms. Solving the above factors can reduce the risk of secondary injury and reduce fatigue false alarms. To achieve this effect, firstly, during assisted training, a fatigue threshold assessment profile corresponding to the user's identifier is obtained, and a dynamic fatigue threshold is generated based on the fatigue threshold assessment profile. Therefore, based on the user's physiological characteristics (such as age and weight), i.e., fatigue threshold assessment profile, a personalized dynamic fatigue threshold, i.e., a personalized fatigue judgment benchmark, can be generated to adapt to individual user differences, avoiding misjudgments caused by fixed thresholds. Next, motion data and electromyographic (EMG) signals are collected in real time and stored in preset motion data queues and EMG signal queues, respectively. This allows for continuous real-time collection of motion data and EMG signals during assisted training, capturing the true muscle responses and actual limb movements. Subsequently, at preset time intervals, motion data sequences and EMG signal sequences are retrieved from the preset motion data queues and EMG signal queues. Then, the aforementioned EMG signal sequences are input into the artifact removal EMG feature extraction module of a pre-trained fatigue monitoring model to obtain EMG artifact removal feature extraction information. The fatigue monitoring model includes an artifact removal EMG feature extraction module, a motion temporal feature extraction module, a spatiotemporal cross-attention fusion module, and an adaptive fatigue state detection module. This removes motion artifacts caused by loose wearable devices, extracts clean EMG signal features, and obtains EMG artifact removal feature extraction information. Next, the aforementioned motion data sequence is input into the aforementioned motion temporal feature extraction module to obtain kinematic temporal feature information. This yields kinematic temporal feature information characterizing kinematic features (e.g., smoothness features during movement). Then, the aforementioned electromyography artifact removal feature extraction information and the aforementioned kinematic temporal feature information are input into the aforementioned spatiotemporal cross-attention fusion module. Feature fusion is performed through a spatiotemporal cross-attention mechanism, enabling deep interaction and fusion of multimodal data to distinguish between "true muscle fatigue" and "motor compensation," avoiding misjudgments caused by single-modal data.Therefore, electromyographic signal features and kinematic features can be fused using a spatiotemporal cross-attention mechanism to obtain motion state coupled feature information. Then, this motion state coupled feature information and a dynamic fatigue threshold are input into the adaptive fatigue state detection module to obtain fatigue detection information. Thus, fatigue state detection can be performed using the fused motion state information and dynamic fatigue threshold, ensuring that fatigue detection can adapt to the individual's real-time state and dynamic threshold, resulting in a more accurate assessment of fatigue state. This avoids the risk of muscle strain or secondary joint injury caused by excessively high thresholds leading to forced training under extreme fatigue, thus reducing the risk of secondary injury. It also reduces the frequency of false fatigue alarms caused by excessively low thresholds. Based on the fatigue detection information, fatigue alerts are generated and broadcast. This allows for timely reminders when the user reaches the fatigue threshold, preventing muscle strain or secondary injury caused by overtraining. Because it employs a combination of dynamic threshold generation (adapting to individual differences), artifact removal (purifying electromyographic signals), and multimodal spatiotemporal cross-attention fusion (distinguishing between fatigue and compensation), it solves the problems of fixed threshold misjudgment, motion artifact contamination, and single-modality limitations. This makes fatigue detection results more accurate, reduces the risk of muscle strain or secondary injury, and also reduces the frequency of fatigue false alarms.

[0076] The above-described embodiments of this disclosure have the following beneficial effects: The rehabilitation training system for motor function reconstruction in spinal cord injury paraplegic patients according to some embodiments of this disclosure reduces motor injury or gait fixation during motor rehabilitation training, shortens the patient's recovery period, and reduces the risk of secondary injury during training. Specifically, the reasons for motor injury or gait fixation, prolonged recovery period, and increased risk of secondary injury during motor rehabilitation training are: the rehabilitation process is fragmented into independent stages, with gaps between different stages. Since the transition from bedside passive training to functional reconstruction training mainly relies on periodic manual assessments by doctors, lacking real-time objective data support, the switching of training modes often lags behind the actual recovery process of the patient's neurological function. If the switch is too early, insufficient residual muscle strength may lead to motor injury or gait fixation; if the switch is too late, the optimal window of neuroplasticity is missed, resulting in low rehabilitation efficiency and prolonged recovery period. Moreover, existing functional reconstruction training systems (such as spinal cord electrical stimulation devices) often adopt an open-loop control mode, that is, outputting electrical pulses according to fixed frequency and parameters, which cannot dynamically adjust stimulation based on real-time changes in the patient's functional state during training, increasing the risk of secondary injury. Based on this, some embodiments of the present disclosure provide a rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia, comprising: a bedside lower limb passive training system, a lower limb function reconstruction and strengthening training system, a server, and a doctor's mobile terminal, wherein: the doctor's mobile terminal is configured to send spinal cord injury information corresponding to a user identifier to the server. Then, the server is configured to perform the following steps: based on the spinal cord injury information, generate first training control information corresponding to the bedside lower limb passive training system. Thus, personalized early first training control information can be generated according to the patient's spinal cord injury information, i.e., the patient's specific injury condition, ensuring safety and targeted intervention in the early stages of rehabilitation. Subsequently, the first training control information is sent to the bedside lower limb passive training system corresponding to the user identifier to control the bedside lower limb passive training device in the system to assist in at least one of the following training methods at a preset frequency: passive joint training, mind-driven assisted movement training, and standing training. Therefore, the bedside lower limb passive training device can be driven at a preset frequency to assist in at least one of the following training methods: passive joint training, mind-driven assisted movement training, and standing training, with at least one targeted training method based on the patient's injury condition. Subsequently, the bedside lower limb passive training device in the aforementioned bedside lower limb passive training system is configured to perform the following steps: First, during mind-driven assisted movement training, a training phase transition detection is performed to generate training phase transition test information. Thus, training phase transition test information can be generated during each mind-driven assisted movement training session to ensure that the patient's potential to enter the next stage can be identified as early as possible.The second step involves sending the aforementioned training phase transition test information to the doctor's mobile terminal, which then forwards the user identifier and training phase transition test information to the server. This allows the server to generate second training control information for entering the next stage of training, namely lower limb function reconstruction and strengthening training. This ensures timely intervention within the optimal window of neural plasticity, reducing the risk of motor injury or gait stagnation due to premature or delayed switching, and prolonging the patient's recovery period. The server is then further configured to generate the second training control information based on the training phase transition test information and send it to the lower limb function reconstruction and strengthening training system corresponding to the user identifier. This allows the system to perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation based on the second training control information. Thus, the lower limb function reconstruction and strengthening training system can perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation, dynamically adjusting stimulation based on real-time changes in the patient's functional state during training, reducing the risk of secondary injury. Because real-time training phase transition detection was performed during each mind-driven assisted movement training session using the bedside lower limb passive training system, training phase transition test information reflecting the patient's true neurological recovery status was generated. Based on this training phase transition test information, the server dynamically generated secondary training control information and controlled the lower limb function reconstruction and strengthening training system to perform closed-loop spinal cord stimulation gait regulation. This eliminated the lag in rehabilitation phase transitions, ensuring timely intervention during the optimal window of neuroplasticity and reducing problems such as motor injury or abnormal gait solidification caused by premature or late switching, thus prolonging the patient's recovery period. Furthermore, by performing at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation and sensing changes in status, the risk of secondary injury during motor function reconstruction training was reduced.

[0077] Figure 2 The flowchart 200 illustrates some embodiments of a method for transmitting training phase transition test information using a bedside lower limb passive training device included in the rehabilitation training system for motor function reconstruction in spinal cord injury paraplegic patients according to the present disclosure. This method for transmitting training phase transition test information includes the following steps: Step 201: Perform training phase transition detection in mind-driven assisted movement training to generate training phase transition test information.

[0078] In some embodiments, the execution subject of the training phase transition test information sending method (e.g., a bedside lower limb passive training device) can perform training phase transition detection during mind-driven assisted movement training to generate training phase transition test information.

[0079] Step 202: Send the training phase conversion test information to the doctor's mobile terminal so that the doctor's mobile terminal can send the user identifier and training phase conversion test information to the server.

[0080] In some embodiments, the execution entity may send the training phase transition test information to the doctor's mobile terminal, so that the doctor's mobile terminal may send the user identifier and the training phase transition test information to the server.

[0081] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of technical features, but should also cover other technical solutions formed by arbitrary combinations of technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A rehabilitation training system for the reconstruction of motor function in patients with spinal cord injury and paraplegia, comprising: Bedside lower limb passive training system, lower limb function reconstruction and strengthening training system, server, doctor's mobile terminal, among which: The doctor's mobile terminal is configured to send spinal cord injury information corresponding to the user's identifier to the server; The server is configured to perform the following steps: Based on the spinal cord injury information, first training control information corresponding to the bedside lower limb passive training system is generated; The first training control information is sent to the bedside lower limb passive training system corresponding to the user identifier, so as to control the bedside lower limb passive training device in the bedside lower limb passive training system to assist in at least one of the following training at a preset frequency: passive joint training, mind-driven assisted movement training and standing training. The bedside lower limb passive training device in the bedside lower limb passive training system is configured to perform the following steps: Training phase transition detection is performed in mind-driven assisted movement training to generate training phase transition test information; The training phase transition test information is sent to the doctor's mobile terminal, so that the doctor's mobile terminal can send the user identifier and the training phase transition test information to the server; The server is further configured to generate second training control information based on training phase transition test information, and to send the second training control information to the lower limb function reconstruction and strengthening training system corresponding to the user identifier, so that the lower limb function reconstruction and strengthening training system can perform at least one motor function reconstruction training under closed-loop spinal cord stimulation gait regulation based on the second training control information.

2. The rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia according to claim 1, wherein, The bedside lower limb passive training system includes a bedside lower limb passive training device, an external decoder, and a surface electromyography electrode array. The bedside lower limb passive training device is further configured to perform training phase transition detection during mind-driven assisted movement training through the following steps to generate training phase transition test information: In each mind-driven assisted movement training session, the system receives the EEG decoding confidence score from an external decoder and collects leg electromyography (EMG) signals through a surface EMG electrode array. The EEG decoding confidence score and leg EMG signals are then stored in the EEG decoding confidence score queue and the leg EMG signal queue, respectively. Based on the EEG decoding confidence queue and the leg electromyography signal queue, training device conversion prompt information is generated and sent to the doctor's mobile terminal. In response to the detection of a training device conversion prompt, the device works in conjunction with the suspended weight reduction equipment to perform the test task. When executing a test task, perform the following test steps: By collecting brainwave signals generated by the discharge of nerve cell groups in the user's motor cortex and supplementary motor area through intravascular EEG electrodes, a potential amplitude sequence is obtained. Electromyographic signal sequences were acquired using a surface electromyographic electrode array; The inertial measurement unit collects inertial data for each moving part. Pressure information for the left foot and the right foot was collected using a pressure sensor array. Based on the potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group and right foot pressure information group, training phase transition test information is generated.

3. The rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia according to claim 2, wherein, Each potential amplitude in the potential amplitude sequence corresponds to a sampling time, each electromyographic signal in the electromyographic signal sequence corresponds to a sampling time, and the bedside lower limb passive training device is further configured to generate training phase transition test information based on the potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group, and right foot pressure information group through the following steps, including: The potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group and right foot pressure information group are input into the modality-specific feature extraction layer of the pre-trained rehabilitation assessment model to obtain potential feature information, electromyographic feature information, kinematic feature information and dynamic feature information. The rehabilitation assessment model includes the modality-specific feature extraction layer, feature fusion and interaction layer and functional index regression layer. The potential feature information, the electromyographic feature information, the kinematic feature information, and the dynamic feature information are fed into the feature fusion and interaction layer to obtain cross-modal joint behavioral feature information; The cross-modal joint behavioral feature information is input into the functional index regression layer to obtain the comprehensive walking ability level, neuromotor control score, weight-bearing propulsion score and gait stability score; The comprehensive walking ability level, the neuromotor control score, the weight-bearing propulsion score, and the gait stability score are determined as the training phase transition test information.

4. The rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia according to claim 3, wherein, The bedside lower limb passive training device is further configured to input the potential amplitude sequence, electromyographic signal sequence, various inertial data, left foot pressure information group, and right foot pressure information group into the modality-specific feature extraction layer of a pre-trained rehabilitation assessment model through the following steps, to obtain potential feature information, electromyographic feature information, kinematic feature information, and dynamic feature information, including: The potential amplitude sequence is input into the potential feature extraction subnetwork included in the modality-specific feature extraction layer to obtain potential feature information. The modality-specific feature extraction layer includes the potential feature extraction subnetwork, electromyographic feature extraction subnetwork, kinematic feature extraction subnetwork, and pressure feature extraction subnetwork. The electromyographic signal sequence is input into the electromyographic feature extraction subnetwork to obtain electromyographic feature information; The inertial data are input into the kinematic feature extraction subnetwork to obtain kinematic feature information; The pressure information groups of the left foot and the right foot are input into the pressure feature extraction subnetwork to obtain dynamic feature information.

5. The rehabilitation training system for motor function reconstruction in patients with spinal cord injury and paraplegia according to claim 1, wherein, The training phase transition test information includes a comprehensive walking ability level, a neuromotor control score, a weight-bearing propulsion score, and a gait stability score. The lower limb function reconstruction and strengthening training system includes a weight-reducing walking training device, a gait rehabilitation robot, and an active training lower limb exoskeleton robot. The server is further configured to generate second training control information based on the training phase transition test information through the following steps: In response to determining that the overall walking ability level is greater than or equal to the first preset level and less than or equal to the second preset level, the following steps are performed: In response to the fact that the neuromotor control score included in the training phase transition test information is less than the preset neuromotor control score, the preset parameter control information corresponding to the active training type lower limb exoskeleton robot is determined as the second training control sub-information. In response to the determination that the weighted propulsion score included in the training phase transition test information is less than the preset weighted propulsion score, the preset parameter control information corresponding to the weight-reducing walking training device is determined as the second training control sub-information. In response to the determination that the gait stability score included in the training phase transition test information is less than the preset gait stability score, the preset parameter control information of the gait rehabilitation robot is determined as the second training control sub-information; At least one of the determined second training control sub-information is identified as the second training control information.

6. A method for sending training phase transition test information, applied to the bedside lower limb passive training device included in the rehabilitation training system for motor function reconstruction of spinal cord injury paraplegic patients as described in any one of claims 1-6, the method comprising: Training phase transition detection is performed in mind-driven assisted movement training to generate training phase transition test information; The training phase transition test information is sent to the doctor's mobile terminal, so that the doctor's mobile terminal can send the user identifier and the training phase transition test information to the server.

Images