A lower limb rehabilitation monitoring and evaluation system and method based on multi-modal sparse data

Abstract

This invention discloses a lower limb rehabilitation monitoring and assessment system and method based on multimodal sparse data, belonging to the fields of rehabilitation medicine engineering and wearable sensing technology. The system includes a hardware terminal and host computer software. The hardware terminal includes an EMG acquisition circuit, an IMU module circuit, a power supply module, and a microcontroller. The host computer software includes a Bluetooth communication module, a data receiving and buffering module, an algorithm implementation module, and a UI layout module. The hardware terminal adopts a sparse wearable structure, with the IMU circuit board fixed to the anterior side of the tibia and the EMG lead wires extending to the gastrocnemius muscle. This invention simultaneously acquires EMG and IMU data from both the healthy and affected sides, calculates gait frequency, support phase ratio, muscle exertion level, fatigue level, and symmetry index through multimodal fusion, and automatically generates a gait analysis report. This invention solves the problem of conflicting EMG and IMU wearing positions and has the advantages of low cost and home use.

Description

Technical Field

[0001] This invention relates to the fields of rehabilitation medicine engineering and wearable sensing technology, and in particular to a lower limb rehabilitation monitoring and assessment system and method based on multimodal sparse data. Background Technology

[0003] Currently, lower limb gait assessment mainly relies on camera-based image analysis and wearable sensor-based data acquisition and analysis. Camera methods can directly collect posture data, but they are costly, require fixed locations, and are cumbersome to operate, making them difficult to deploy in home environments. Furthermore, they cannot capture muscle and neurophysiological activity. Among wearable sensors, surface electromyography (EMG) signals can capture muscle activation timing and contraction intensity in real time, while inertial measurement units (IMUs) can calculate kinematic parameters such as joint angles and gait cycles. However, single-sensor data is limited: EMG is susceptible to motion artifacts and cannot capture joint movement trajectories; IMUs can only detect limb movements and cannot reflect muscle and neuromuscular control mechanisms. Neither can independently complete a comprehensive assessment.

[0004] Most existing research relies on a single modality, while a few attempts at multimodal fusion systems, such as commercial products from Noraxon and Cometa, achieve simultaneous EMG and IMU data acquisition. However, these systems are costly, cumbersome to wear, and primarily geared towards professional laboratories. The fundamental bottleneck lies in the physical conflict between the EMG sensor's close contact with the target muscle and the IMU's fixation at the bone joint. When integrating these two systems, a compromise must be made, resulting in a decrease in the data quality of at least one sensor. Therefore, current technologies consistently struggle to achieve low-cost, high-quality, and home-based multimodal lower limb gait assessment suitable for rehabilitation. Summary of the Invention

[0005] This invention addresses the problems of physical conflicts between EMG and IMU in terms of wearing position and low system integration in the prior art, as well as the lack of single-modal information dimension, by providing a lower limb rehabilitation monitoring and assessment system and method based on multimodal sparse data.

[0006] A lower limb rehabilitation monitoring and assessment system based on multimodal data fusion includes a hardware terminal and host computer software, wherein the hardware terminal and the host computer software communicate wirelessly via Bluetooth.

[0007] The hardware terminal includes: an EMG acquisition circuit for acquiring surface electromyographic signals of key muscle groups in the lower limbs; an IMU module circuit for acquiring acceleration, angular velocity, and Euler angle data of lower limb movement; a power supply module for supplying power to each circuit; and a microcontroller for aggregating EMG and IMU data and sending it to a host computer via Bluetooth.

[0008] The hardware terminal includes: an EMG acquisition circuit for acquiring surface electromyographic signals of key muscle groups in the lower limbs; an IMU module circuit for acquiring acceleration, angular velocity, and Euler angle data of lower limb movement; a power supply module for supplying power to each circuit; and a microcontroller for aggregating EMG and IMU data and sending it to a host computer via Bluetooth.

[0009] The host computer software includes: a Bluetooth communication module for establishing a Bluetooth connection with the hardware terminal and receiving data; a data receiving and buffering module for restoring the Bluetooth byte stream into structured frames and performing field parsing and caching; an algorithm implementation module for calculating gait evaluation indicators based on EMG signals and IMU data; and a UI layout module for real-time display of waveforms, evaluation indicators, and interactive control.

[0010] The EMG acquisition circuit includes, in sequence, an instrumentation amplifier, a right leg drive circuit, a dual-T notch filter, an active bandpass filter, a post-amplifier circuit, an adder, and a voltage follower. The instrumentation amplifier performs preliminary amplification of the raw electromyography signal, with a gain of [value missing]. Approximately 10 times. The right leg drive circuit extracts the average common-mode voltage through a resistor network, amplifies it in reverse, and then sends it back to the human body to suppress power frequency interference.

[0011] Formula for the center frequency of a double-T notch filter: The center frequency is 50 Hz, used to remove power frequency interference. The upper cutoff frequency of the active bandpass filter is... The formula for the lower cutoff frequency is: The passband range is 15~160 Hz. The subsequent amplifier circuit is a non-inverting amplifier circuit with a gain of [missing value]. Approximately 200 times. The adder introduces a DC bias of 1.67V into the signal and boosts it to... The range is provided for acquisition by the microcontroller's ADC. The voltage follower is used to isolate the subsequent circuitry.

[0012] The voltage follower is used to isolate the subsequent circuitry. The hardware terminal adopts a sparse wearable structure, specifically: the overall circuit board containing the IMU module is fixed to the anterior side of the tibia of the lower leg, and the acquisition of EMG signals is extended to the gastrocnemius muscle through lead wires for electromyography signal measurement, realizing the separation and consideration of the wearing positions of the IMU and EMG sensors.

[0013] The hardware terminal includes two independent devices, worn on the patient's healthy and affected lower limbs respectively, to achieve synchronous acquisition of bilateral electromyographic signals and inertial data. The microcontroller ADC is in continuous conversion mode, recording an EMG signal every 2ms. Simultaneously, it receives acceleration, angular velocity, and Euler angle data from the IMU via serial port interrupt. After recording 5 EMG data, the IMU and EMG data are sent to the host computer via Bluetooth. The EMG sampling frequency is 500Hz, and the IMU sampling frequency is 100Hz.

[0014] The Bluetooth module follows the BLE5.3 communication protocol to achieve low-power wireless data transmission.

[0015] The UI layout module is built on the PyQt5 framework and includes a left-side control and information summary panel, a central real-time waveform and muscle status monitoring panel, and a right-side calibration panel. The central waveform display uses the pyqtgraph library to achieve high-frequency real-time refresh of bilateral electromyographic signals, and the underlying layer uses a double-buffered queue to down-convert high-frequency data to 4Hz for graphics rendering. The system uses the QStackedWidget control to switch between different clinical scenarios and uses color-coded thresholds to intuitively reflect the symmetry index to the interface.

[0016] The host computer software also includes a large model interpretation report module, which is used to encapsulate the subject information and multi-dimensional quantitative assessment results after gait quantification is completed. It calls the cloud-based large language model API through structured prompts to automatically generate an assessment report containing gait feature summaries, anomaly analysis, and rehabilitation intervention suggestions. This module is encapsulated as a background asynchronous thread by inheriting from QThread.

[0017] The host computer software also includes a cloud data upload module, which is based on the Supabase cloud database. After the assessment task is completed, the module organizes patient information, quantitative indicators and analysis timestamps into a local structured file and uploads it to a remote object storage through the cloud storage interface. The cloud path is organized in a hierarchical manner of "patient name / timestamp / file name".

[0018] The calculation method for the gait evaluation index in the algorithm implementation module includes:

[0019] Preprocessing of EMG signals:

[0020] Real-time processing of EMG signals:

[0021] DC removal processing:

[0022] Eliminate very low frequency baseline drift;

[0023] Bandpass filtering:

[0024] A 4th-order Butterworth bandpass filter is used:

[0025] Preserve the effective frequency band and filter out out-of-band noise;

[0026] Full-wave rectification: ;

[0027] Calculation of muscle exertion:

[0028] Average absolute value:

[0029] It directly reflects the degree of muscle activation and the magnitude of force exerted, and is positively correlated with the muscle's contractile force;

[0030] Root mean square: This represents the average energy level of the signal and is positively correlated with the muscle activation intensity and force exertion, used to assess the degree of muscle exertion.

[0031] Muscle fatigue level calculation:

[0032] Fuzzy entropy:

[0033] ,

[0034] For the embedded dimension; For similarity threshold, For signal length, express The average similarity of subsequences is used to measure the complexity of time series.

[0035] Calculate the percentage of muscle fatigue:

[0036] Where baseline represents the baseline fuzzy entropy, which is the average fuzzy entropy of the muscle before fatigue at the initial exertion stage. This represents the average fuzzy entropy value at the current moment; it is used to characterize muscle fatigue in real time.

[0037] IMU gait parameter calculation:

[0038] acceleration magnitude: The IMU's three-axis accelerations are combined into a single scalar acceleration to eliminate the influence of direction. A threshold is set to distinguish whether a sampling point is within the walking range; step frequency is calculated in real time.

[0039] ;

[0040] in The number of steps detected during walking. Walking time;

[0041] Real-time calculation of the phase ratio: ,

[0042] in Indicates the number of sampling points in the supporting phase; This represents the total number of sampling points for a complete gait cycle.

[0043] Symmetry index calculation:

[0044] ,

[0045] Where H represents the gait parameters of the healthy side; A represents the gait parameters of the affected side, used to quantify the degree of asymmetry in the gait parameters of the left and right limbs; the closer to 0, the better the symmetry.

[0046] This invention also provides a lower limb rehabilitation monitoring and assessment method based on the above system, comprising the following steps: Step 1: Wearing two sets of hardware terminals on the patient's healthy lower limb and affected lower limb respectively; Step 2: Starting the host computer software and establishing a connection with the two sets of hardware terminals via Bluetooth; Step 3: Performing maximum voluntary contraction MVC calibration to obtain the patient's maximum muscle exertion baseline; Step 4: The patient begins walking, and the hardware terminals collect bilateral EMG signals and IMU data in real time and transmit them to the host computer via Bluetooth; Step 5: The host computer displays bilateral electromyographic waveforms, muscle exertion status, and fatigue status in real time, and calculates gait frequency and support phase ratio in real time; Step 6: After the assessment, calculating the overall assessment index and bilateral symmetry index, and generating a gait analysis report.

[0047] This invention has the following effects:

[0048] (1) The present invention adopts a wearing method in which the IMU circuit board is fixed on the front of the tibia and the EMG acquisition is extended to the gastrocnemius muscle through the lead wire, which solves the problem of signal quality degradation caused by the conflict of wearing positions between EMG and IMU sensors.

[0049] (2) The present invention adopts an assessment paradigm of simultaneous collection and symmetrical analysis of healthy and affected sides, using the patient's own healthy side data as a natural reference benchmark, effectively avoiding the influence of individual differences such as gender, age, and medical history on the assessment results, and intuitively reflecting the progress of rehabilitation through bilateral comparison.

[0050] (3) This invention combines the muscle electrophysiological information of EMG with the kinematic information of IMU through multimodal data fusion analysis, and realizes the synchronous calculation of multi-dimensional indicators such as cadence, support phase ratio, muscle exertion degree, and muscle fatigue degree, overcoming the limitation of insufficient information dimensions of a single sensor.

[0051] (4) This invention introduces cloud-based large model API to automatically generate gait analysis reports. Through structured prompt word engineering, multimodal quantitative indicators are transformed into assessment reports and rehabilitation suggestions that are both professional and readable, effectively solving the pain point that the original physiological parameters are difficult for patients to understand directly. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0053] Figure 2 This is a schematic diagram of the EMG acquisition circuit of the present invention;

[0054] Figure 3 This is a diagram of the microcontroller program structure of the present invention;

[0055] Figure 4 This is a structural diagram of the host computer software module of the present invention;

[0056] Figure 5 This is a user operation flowchart for the present invention;

[0057] Figure 6 The diagram shows the sparse wearable structure of the present invention ((a) is a product diagram; (b) is a wearable diagram).

[0058] Figure 7 This is a schematic diagram of the host computer UI interface of the present invention. Detailed Implementation

[0059] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, and the terms "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0060] Example 1 Hardware Terminal Design

[0061] like Figure 1 As shown, the lower limb rehabilitation monitoring and assessment system based on multimodal data fusion in this embodiment includes a hardware terminal and host computer software. The hardware terminal and the host computer software communicate wirelessly via Bluetooth.

[0062] like Figure 2As shown, the hardware terminal includes an EMG acquisition circuit, an IMU module circuit, a power supply module, and a microcontroller. The power supply module uses a 3.7V rechargeable lithium battery, converting the 3.7V voltage to ±5V and 3.3V to supply subsequent circuits. The IMU module uses a JY901s module, communicating with the microcontroller via a serial port to collect data such as acceleration, angular velocity, and Euler angles during walking. The Bluetooth module uses an MX01, conforming to the BLE5.3 communication protocol. The microcontroller uses an STM32F103C8T6, which aggregates the EMG and IMU data and sends it to the host computer via the Bluetooth module.

[0063] like Figure 3 As shown, the microcontroller mainly uses an ADC and a serial port peripheral. The ADC is in continuous conversion mode. After receiving the acquisition command from the host computer, the main function records the EMG signal every 2ms, while simultaneously receiving acceleration, angular velocity, and Euler angle data from the IMU via a serial port interrupt. After recording a total of 5 EMG data, the IMU and EMG data are sent to the host computer via Bluetooth serial port. The EMG sampling frequency is 500Hz, and the IMU sampling frequency is 100Hz.

[0064] like Figure 6 As shown, Figure 6 As shown in (a), the hardware terminal includes a main circuit board integrating an IMU module and a microcontroller, as well as EMG acquisition electrodes connected via lead wires. Figure 6 As shown in (b), the main circuit board is fixed to the anterior side of the tibia of the lower leg by a strap, and the EMG acquisition electrode is fixed to the gastrocnemius muscle belly on the same side by a patch. The hardware terminal adopts a sparse wearable structure, specifically: the overall circuit board containing the IMU module is fixed to the anterior side of the tibia of the lower leg, and the acquisition of EMG signals extends to the gastrocnemius muscle through lead wires to measure electromyographic signals, realizing the separation and consideration of the wearing positions of the IMU and EMG sensors. The hardware terminal includes two independent devices, worn on the patient's healthy lower limb and affected lower limb respectively, to realize the synchronous acquisition of bilateral electromyographic signals and inertial data.

[0065] The EMG acquisition circuit includes an instrumentation amplifier INA128, a right leg drive circuit, a dual-T notch filter, an active bandpass filter, a post-amplifier circuit, an adder, and a voltage follower.

[0066] The raw signal was initially amplified using an INA128 instrumentation amplifier, with a gain of [missing value]. =9.93, approximately 10 times. Simultaneously, the right leg drive circuit extracts the average common-mode voltage through a resistor network, amplifies this signal in reverse, and then transmits it back to the human body, thereby suppressing power frequency interference.

[0067] The initially amplified signal first passes through a double-T notch filter to remove power frequency interference. The center frequency formula for the double-T notch filter is used as follows: The notch filter needs to have a center frequency of 50 Hz, therefore a resistor R = 31.6kΩ and a capacitor C = 100nF are selected. The signal after notching passes through an active bandpass filter to remove other high-frequency and low-frequency noise. The full bandwidth of the electromyography signal is 0-500Hz, and the bandpass filter is designed with a passband of 15-160Hz and an upper cutoff frequency. Lower cutoff frequency The overall passband is 15.9-157.79Hz.

[0068] The subsequent amplification stage uses a non-inverting amplifier circuit with a gain of [missing information]. With the addition of the pre-amplifier, the overall circuit gain is 1680 times. To enable signal acquisition using a microcontroller-based ADC, the signal needs to be boosted to the 0-3.3 V range. The adder introduces a 1.67 V DC bias, boosts the signal to the 0-3.3 V range, and finally isolates it from the subsequent circuitry via a voltage follower. The average output signal of the adder is 1.596 V, meeting the ADC acquisition requirements. The common-mode rejection ratio is... .

[0069] Example 2: Host Computer Software Design

[0070] like Figure 4 As shown, the host computer software includes a Bluetooth communication module, a data receiving and buffering module, an algorithm implementation module, a UI layout module, a large model interpretation report module, and a cloud data upload module.

[0071] The Bluetooth communication module, using PyQt5 and an asynchronous Bluetooth library, enables dual-channel hardware concurrent connections, data frame splicing and parsing, and level conversion. It is responsible for establishing, maintaining, and restoring Bluetooth connections between the host computer and the left and right devices. The data receiving and buffering module is responsible for restoring the Bluetooth byte stream into structured frames, performing field parsing, unit conversion, side-by-side splitting, and multi-level buffering. The algorithm implementation module performs calculations of core biomechanical indicators, including electromyography signal preprocessing, MAV / RMS features, fuzzy entropy fatigue estimation, gait frequency event detection, and support phase ratio estimation.

[0072] like Figure 5 As shown, the UI module is responsible for interface organization, interaction control, and result visualization. Figure 7As shown, the UI module is built on the PyQt5 framework and adopts a modern medical-style borderless card-based interface design. The main interface is divided into a left-side control and information summary panel, a middle-side real-time waveform and muscle status monitoring panel, and a right-side calibration panel. The waveform display in the middle is implemented based on the pyqtgraph library, and the underlying layer uses a double-buffered queue to down-convert high-frequency data to 4 Hz for graphics rendering. The system uses the QStackedWidget control to switch between different clinical business scenarios, and automatically calls a symmetry algorithm to calculate and normalize multidimensional indicators after data acquisition.

[0073] In addition, a mobile app was designed based on the Flutter development framework, including a device connection interface, a real-time data display interface, an evaluation result interface, and a data management interface.

[0074] After gait quantification data collection, the large-scale model interpretation report module encapsulates the subject's basic information and quantitative assessment results. Using a structured prompt template, it calls the cloud-based large language model API across threads to automatically generate an assessment report containing gait feature summaries, anomaly analysis, and rehabilitation intervention suggestions. The cloud data upload module, based on the Supabase cloud database, organizes patient information, quantitative indicators, and analysis timestamps into a local structured file and uploads it to the cloud-based remote object storage after the assessment task is completed.

[0075] Example 3: Calculation of Gait Evaluation Indicators

[0076] The specific calculation method for the gait evaluation index in the algorithm implementation module is as follows:

[0077] For EMG signals, DC removal processing is performed: Eliminate very low frequency baseline drift; employ a 4th-order Butterworth bandpass filter: Preserve effective frequency band and filter out out-of-band noise; full-wave rectification: ;

[0078] Calculation of muscle exertion: ; .

[0079] Muscle fatigue level calculation: fuzzy entropy ,

[0080] For the embedded dimension; For similarity threshold, For signal length, express The average similarity of subsequences is used to measure the complexity of a time series; when muscles are fatigued, the complexity of EMG signals decreases, and the fuzzy entropy decreases significantly.

[0081] The percentage of muscle fatigue (fatigue%) converts the change in fuzzy entropy into a fatigue percentage of 0 to 100%. Where baseline represents the baseline fuzzy entropy, which is the average fuzzy entropy of the muscle before fatigue at the initial exertion stage. This represents the average fuzzy entropy value at the current moment; it is used to characterize muscle fatigue in real time.

[0082] IMU gait parameter calculation: Acceleration magnitude: By setting a threshold, walking intervals can be distinguished;

[0083] Cadence is the number of steps taken per minute while walking. ;

[0084] in The number of steps detected during walking. Walking time; Real-time calculation of support phase ratio: ,in Indicates the number of sampling points in the supporting phase; This represents the total number of sampling points for a complete gait cycle.

[0085] Symmetry index calculation: Where H represents the gait parameters of the healthy side; A represents the gait parameters of the affected side, used to quantify the degree of asymmetry in the gait parameters of the left and right limbs; the closer to 0, the better the symmetry.

[0086] Example 4 System Testing and Verification

[0087] Hardware test results show that the pre-amplifier gain is 10x, the notch depth is -17.4 dB, the bandpass filter passband is 15~160 Hz, the overall circuit gain is 1680x, and the common-mode rejection ratio is 94.46 dB. In the overall EMG acquisition circuit test, the calf muscle was selected. When the muscle was relaxed, the output maintained a baseline; when the calf contracted forcefully, the oscilloscope immediately showed a clear signal, demonstrating its ability to sensitively capture muscle changes.

[0088] Embedded software testing shows that the microcontroller and host computer can communicate normally via Bluetooth, with normal data format and no packet loss issues. The EMG sampling frequency of 500 Hz meets the Nyquist sampling requirements, and the IMU sampling frequency of 100 Hz meets the gait analysis requirements. Using normal human walking EMG and IMU data to test the algorithm's functionality, the MAV and RMS values ​​show obvious periodicity, and the state classification diagram conforms to walking patterns.

[0089] During the overall system test, the subject walked normally, and the host computer plotted the electromyographic signals of the gastrocnemius muscles of both legs in real time in the central area, while the lower left corner calculated the cadence and the stance phase of each leg in real time. After the evaluation, the system calculated the overall indicators and the symmetry of the left and right legs based on the data from the entire process.

[0090] Data collected from patients and healthy individuals during walking were compared and analyzed. The results are shown in Tables 1 and 2.

[0091] Table 1. Comparison of gait support between healthy individuals and patients.

[0092] Table 2. Analysis of RMS indices for the left and right legs in healthy individuals and patients.

[0093] Table 1 shows the IMU gait data collected from patients and healthy individuals during walking, and the results of analyzing the support phase indices of the two groups. The comparison results show that the symmetry between the two sides of patients is mostly above 0.1, while the indices of healthy individuals are mostly below 0.1. Furthermore, the proportion of the support phase during walking is significantly higher on the affected side than on the healthy side in patients. Therefore, using support phase analysis to compare the healthy and affected sides has certain value in rehabilitation assessment.

[0094] Table 2 shows the analysis results of RMS index calculated based on EMG signals from patients and healthy individuals throughout the entire assessment process. It is evident that there are significant differences in RMS indices between patients and healthy individuals. A clear asymmetry exists between the RMS indices of the patient's left and right legs, with the RMS index on the healthy side being significantly higher than that on the affected side, reflecting insufficient strength in the affected limb. This is consistent with the characteristics of hemiplegia in stroke patients. However, as rehabilitation progresses, the function of the affected limb will recover, and the two sides will become increasingly symmetrical. Therefore, this index can be used as a basis for evaluating the patient's rehabilitation effect.

[0095] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features.

Claims

1. A lower limb rehabilitation monitoring and assessment system based on multimodal sparse data, characterized in that, The system includes a hardware terminal and host computer software, with the hardware terminal and host computer software communicating wirelessly via Bluetooth. The hardware terminal includes: an EMG acquisition circuit for acquiring surface electromyography signals of key lower limb muscle groups; an IMU module circuit for acquiring acceleration, angular velocity, and Euler angle data of lower limb movement; a power supply module for supplying power to each circuit; and a microcontroller for aggregating EMG and IMU data and sending it to the host computer via the Bluetooth module. The host computer software includes: a Bluetooth communication module for establishing a Bluetooth connection with the hardware terminal and receiving data; a data receiving and buffering module for restoring the Bluetooth byte stream into structured frames and performing field parsing and buffering; an algorithm implementation module for calculating gait evaluation indicators based on EMG signals and IMU data; and a UI layout module for real-time display of waveforms, evaluation indicators, and interactive control.

2. The lower limb rehabilitation monitoring and assessment system based on multimodal data sparsity according to claim 1, characterized in that, The EMG acquisition circuit includes, in sequence, an instrumentation amplifier, a right leg drive circuit, a dual-T notch filter, an active bandpass filter, a post-amplifier circuit, an adder, and a voltage follower; the gain formula of the instrumentation amplifier is: ； Formula for the center frequency of a double-T notch filter: ； The formula for the upper cutoff frequency of an active bandpass filter is: ； The formula for the lower cutoff frequency is: ； The gain formula for the subsequent amplifier circuit is: The adder introduces the signal into the DC bias and raises it to the 0~3.3V range for acquisition by the microcontroller's ADC.

3. The lower limb rehabilitation monitoring and assessment system based on multimodal data fusion according to claim 1, characterized in that, The hardware terminal adopts a split-type sparse wearable structure, specifically: the main circuit board containing the IMU module and the microcontroller is fixed to the anterior tibial bone surface of the lower leg by a strap; the EMG acquisition electrode is separately set from the main circuit board by a flexible lead wire and fixed to the gastrocnemius muscle belly on the same side by a patch; the linear distance between the main circuit board and the EMG acquisition electrode is not less than 50mm.

4. The lower limb rehabilitation monitoring and assessment system based on multimodal data fusion according to claim 1, characterized in that, The hardware terminal includes two independent devices, which are worn on the patient's healthy lower limb and affected lower limb respectively, to achieve synchronous acquisition of bilateral electromyographic signals and inertial data; the algorithm implementation module calculates the symmetry index based on bilateral data and uses the patient's own healthy side data as a reference benchmark to evaluate the function of the affected side.

5. The lower limb rehabilitation monitoring and assessment system based on multimodal data fusion according to claim 1, characterized in that, The UI layout module is built on the PyQt5 framework and includes a left control and information summary panel, a middle real-time waveform and muscle status monitoring panel, and a right calibration panel. The waveform display in the middle section uses the pyqtgraph library to achieve high-frequency real-time refresh of bilateral electromyographic signals, and the underlying layer uses a double-buffered queue to down-rate the high-frequency data to the graphics rendering frame rate.

6. The lower limb rehabilitation monitoring and assessment system based on multimodal data fusion according to claim 1, characterized in that, It also includes a large model interpretation report module, which encapsulates subject information and multi-dimensional quantitative assessment results after gait quantification data collection. It calls the cloud-based large language model API through structured prompts to automatically generate an assessment report containing gait feature summaries, anomaly analysis, and rehabilitation intervention suggestions. This module is encapsulated as a background asynchronous thread by inheriting from QThread.

7. The lower limb rehabilitation monitoring and assessment system based on multimodal data fusion according to claim 1, characterized in that, It also includes a cloud data upload module, which is used to organize patient information, quantitative indicators and analysis timestamps into local structured files after the assessment task is completed, and upload them to remote object storage through the cloud storage interface.

8. A method for monitoring and assessing lower limb rehabilitation based on the system described in any one of claims 1-7, characterized in that, The procedure includes the following steps: Step 1: Wear two sets of hardware terminals on the patient's healthy and affected lower limbs respectively; Step 2: Start the host computer software and establish a connection with the two hardware terminals via Bluetooth; Step 3: Perform maximum voluntary contraction MVC calibration to obtain the patient's maximum muscle exertion baseline; Step 4: The patient begins walking, and the hardware terminals collect bilateral EMG signals and IMU data in real time and transmit them to the host computer via Bluetooth; Step 5: The host computer displays bilateral electromyographic waveforms, muscle exertion status, and fatigue status in real time, and calculates gait frequency and support phase ratio in real time; Step 6: After the assessment, calculate the overall assessment indicators and bilateral symmetry index, and generate a gait analysis report.

9. The method for monitoring and assessing lower limb rehabilitation based on multimodal data fusion according to claim 8, characterized in that, The specific methods for real-time calculation of step frequency and support ratio in step 5 include: Step 51: Real-time processing of EMG signals: Step 511: Perform DC removal processing on the acquired EMG signals sequentially; Step 512: Calculate the mean absolute value and root mean square value to represent the degree of muscle exertion in real time; Step 513: Calculate the fuzzy entropy: And calculate the percentage of muscle fatigue: Where baseline represents the baseline fuzzy entropy, This represents the average fuzzy entropy value at the current moment; it is used to characterize muscle fatigue in real time. Step 52: Real-time processing of IMU data: Step 521: Calculate the resultant acceleration magnitude based on IMU triaxial acceleration: By setting a threshold, it can distinguish whether the sampling point is in the walking range and detect gait events; Step 522: Calculate step frequency in real time; Step 523: Real-time calculation of the support phase ratio: ,in Indicates the number of sampling points in the supporting phase; This represents the total number of sampling points for a complete gait cycle.

10. A method for monitoring and assessing lower limb rehabilitation based on multimodal data fusion according to claim 8, characterized in that, The specific methods for calculating the overall evaluation index and the bilateral symmetry index in step 6 include: Step 61: Based on the EMG signals throughout the entire assessment period, calculate the mean absolute value (MAV) and root mean square value (RMS) of each side as an overall assessment index of the degree of muscle exertion on both sides. Step 62: Based on the IMU data throughout the entire evaluation period, calculate the average cadence and average support phase ratio of each side as the overall evaluation index of bilateral gait stability. Step 63: Calculate the symmetry index of the healthy and affected sides for each of the above indicators: Where H represents the gait parameters of the healthy side; A represents the gait parameters of the affected side, used to quantify the degree of asymmetry in the gait parameters of the left and right limbs; Step 64: Summarize all the calculated overall evaluation indicators and symmetry indices to generate a gait analysis report; Step 65: After the assessment, organize the patient information, key quantitative indicators and analysis timestamps into a local structured file and upload it to the remote object storage via the cloud storage interface.

Images