Orthopedic rehabilitation monitoring network system and method based on edge intelligent decision technology

Abstract

This invention relates to the field of edge computing technology for medical monitoring, specifically to an orthopedic rehabilitation monitoring network system and method based on edge intelligent decision-making technology. The method discloses an orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology. This method acquires standardized multimodal sensory data from patients, and a local edge computing node calls a built-in lightweight skeletal muscle biomechanical model for real-time calculation, extracting joint range of motion, electromyographic envelope, and gait spatiotemporal parameters. These parameters are then input locally into a pre-trained posture anomaly detection decision tree for frame-by-frame comparison and analysis. When any parameter deviates from a preset standard, a posture anomaly event label and associated sensor data slice are immediately generated locally. This lightweight alarm information is transmitted to a regional rehabilitation monitoring center server, triggering the retrieval of high-definition images. This method achieves local real-time intelligent detection and early warning of rehabilitation anomalies, reducing network load and response latency.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology for medical monitoring, specifically to an orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology. Background Technology

[0002] Traditional orthopedic rehabilitation monitoring primarily relies on regular outpatient assessments or simple wearable device monitoring. These devices typically only collect raw sensor data and stream large amounts of raw data to the cloud or central server for processing and analysis via wireless networks. Existing solutions place complex biomechanical calculations and anomaly analysis entirely in a remote center, resulting in significant delays in response from data generation to risk warnings. Abnormal postures and muscle overloads during patient rehabilitation training cannot be identified in real time, posing safety risks. Furthermore, the continuous uploading of raw multimodal data places enormous pressure on network bandwidth and the computational load on the central server, limiting the scalable deployment of monitoring systems.

[0003] This invention aims to address how to achieve real-time calculation of professional-grade biomechanical indicators on resource-constrained terminals, replacing simple data collection and forwarding. It also aims to solve the problem of how to perform real-time decision-making through multi-dimensional feature fusion at the data source without relying on network backhaul and cloud computing power, generating alert events with clear semantics, thereby transforming the monitoring mode from post-event tracing to real-time intervention. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an orthopedic rehabilitation monitoring network system and method based on edge intelligent decision-making technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology, comprising: Standardized multimodal sensing data is acquired, and the local edge computing node calls the built-in lightweight skeletal muscle biomechanical model to perform real-time calculation on the standardized multimodal sensing data frame, extracting the angle change sequence reflecting joint range of motion, the electromyographic envelope signal reflecting muscle load, and the set of spatiotemporal parameters reflecting gait cycle. In the local edge computing node, the angle change sequence of the joint range of motion, the electromyographic envelope signal, and the spatiotemporal parameter set of the gait cycle are input into a pre-trained posture anomaly detection decision tree for frame-by-frame comparison and analysis. When the angle change sequence of the joint range of motion exceeds the safe threshold range, or the electromyographic envelope signal shows an unexpected burst mode, or the spatiotemporal parameter set of the gait cycle deviates from the standard template, posture anomaly event labels and associated sensor data slices are generated. The lightweight alarm information, which includes the posture abnormality event tag and associated sensor data slices, is transmitted to the regional rehabilitation monitoring center server via a wireless body area network gateway. The regional rehabilitation monitoring center server is used to receive and persistently store the lightweight alarm information from multiple local edge computing nodes, and at the same time triggers a process to retrieve high-definition images from the video monitoring equipment of the specified patient.

[0006] As a further aspect of the present invention, a multimodal data sensing network is deployed in the patient rehabilitation area. The multimodal data sensing network consists of a wearable motion capture device, an embedded mattress pressure sensing array, and indoor positioning anchors, and is used to concurrently collect raw streaming data of the patient's skeletal muscle movement trajectory, body pressure distribution cloud map, and indoor activity path. Local edge computing nodes are deployed within the rehabilitation area to aggregate the raw streaming data collected by the wearable motion capture device, the embedded mattress pressure sensing array, and the indoor positioning anchor points to the local edge computing nodes in real time. The local edge computing nodes perform timestamp synchronization and data packet verification on the aggregated raw streaming data to form standardized multimodal sensing data frames.

[0007] As a further aspect of the present invention, inertial measurement data from the wearable motion capture device and pressure distribution matrix from the embedded mattress pressure sensing array are separated from the standardized multimodal sensing data frame. Using the human skeletal linkage model and forward kinematics algorithm defined in the lightweight skeletal muscle biomechanical model, the pose calculation is performed on the inertial measurement data, and the real-time rotation angle and translation amount of the main human joints in three-dimensional space are output, forming the angular change sequence of the joint range of motion. The pressure distribution matrix is ​​subjected to centroid projection and pressure center trajectory tracking calculation. Combined with the coordinates of the foot and hip joints of the human skeletal linkage model, the movement path of the foot pressure center and the offset of the body pressure center in sitting and lying postures are calculated. By integrating the angular change sequence of the joint range of motion with the movement path of the plantar pressure center, and using the gait phase segmentation logic pre-set in the model, continuous gait cycles are divided, and stride length, stride speed, and gait symmetry indicators are extracted from each gait cycle to form a set of spatiotemporal parameters for the gait cycle. The raw electromyographic signal channels in the inertial measurement data are filtered, rectified, and subjected to moving average processing to generate a smooth muscle activation level curve, which serves as the electromyographic envelope signal.

[0008] As a further aspect of the present invention, the pre-trained posture anomaly detection decision tree contains multiple decision nodes defined by rehabilitation medicine experts, and each decision node corresponds to a specific safety threshold range of kinematic or physiological parameters and logical judgment rules. From the angular change sequence of the joint range of motion, extract the flexion angle, extension angle and internal / external rotation angle of the target joint at the current moment, and compare them with the safety threshold range of the corresponding joint in the decision tree; For the electromyographic envelope signal, calculate its root mean square value and peak value in the current gait cycle or movement cycle, and compare it with the expected muscle activation pattern preset for the movement in the decision tree. For the set of spatiotemporal parameters of the gait cycle, calculate the degree of difference between the current gait cycle and the standard rehabilitation stage template gait in terms of stride length and gait symmetry indicators; When any comparison result triggers the abnormal logic set in the decision tree, the current frame data is determined to be abnormal, and the parameter type, actual value, and degree of exceeding the threshold range that triggered the abnormality are packaged to generate the attitude abnormality event label. At the same time, all sensor data within the set time window before and after the abnormality is triggered are cached to form the associated sensor data slice.

[0009] As a further aspect of the present invention, during a normal rehabilitation training cycle in which the abnormal posture event label is not triggered, the local edge computing node calculates the motion smoothness score, motion completion score, and endurance performance score of the current training action based on the angle change sequence of the joint range of motion and the spatiotemporal parameter set of the gait cycle. The calculated motion smoothness score, movement completion score, and endurance performance score are compared with the patient's personal best historical score record and the preset target value of the rehabilitation plan. Based on the comparison results, select the corresponding encouraging voice commands, standard action key reminder voice commands, or intensity adjustment suggestion voice commands from the preset voice prompt library; The local voice broadcasting device connected to the local edge computing node broadcasts the selected voice commands to the patient, guiding the patient to adjust the current or subsequent training actions.

[0010] As a further aspect of the present invention, a second-order difference calculation is performed on the angular change sequence of the joint range of motion to obtain a joint angular acceleration sequence, the number of jitters exceeding a preset smoothing threshold in the joint angular acceleration sequence is counted, and the motion smoothness score is evaluated based on the number of jitters. The joint range of motion angle change sequence is dynamically time-warped and matched with the target angle sequence in the standard rehabilitation movement library. The minimum alignment distance between the sequences is calculated, and the movement completion score is evaluated based on the minimum alignment distance. The endurance performance score is evaluated based on the proportion of time during which the stride length and gait symmetry index in the spatiotemporal parameter set of the gait cycle remain within the acceptable range during the current continuous training period.

[0011] As a further aspect of the present invention, the regional rehabilitation monitoring center server receives and stores the lightweight alarm information from multiple local edge computing nodes corresponding to multiple patients, forming a rehabilitation abnormal event log library; Periodically perform data mining on the rehabilitation abnormal event log database to analyze the statistical patterns and common characteristics of the posture abnormal event tags in patient groups at different rehabilitation stages and with different surgical types; Based on the statistical patterns and common features discovered, the safety threshold range and logical judgment rules in the pre-trained posture anomaly detection decision tree are optimized and updated to generate an updated posture anomaly detection decision tree configuration file. The updated posture anomaly detection decision tree configuration file is distributed to the local edge computing node corresponding to each patient, replacing the original pre-trained posture anomaly detection decision tree, thereby realizing the networked iteration of monitoring rules. The periodic data mining of the rehabilitation abnormal event log database specifically includes: Extract all recorded postural abnormality event tags and their detailed parameter information, patient rehabilitation stage metadata, and surgical type metadata from the rehabilitation abnormality event log library; Cluster analysis was used to group the abnormal posture event labels and identify the categories of abnormal events that occur frequently in patients at specific rehabilitation stages or with specific surgical types. For each type of high-frequency abnormal event, backtrack and analyze the corresponding associated sensor data slices to extract motion pattern features within a set time period before the abnormality occurred. The extracted movement pattern features are compared with the corresponding features in normal rehabilitation data to construct a risk prediction feature vector that can provide early warning of potential abnormalities. The risk prediction feature vector and its corresponding abnormal event category are used as new judgment rules or threshold adjustment criteria and integrated into the updated attitude anomaly detection decision tree configuration file.

[0012] As a further aspect of the present invention, the skin conductivity level signal and heart rate variability signal of the patient are simultaneously collected by integrating a skin conductance sensor and a heart rate sensor into the wearable motion capture device. The local edge computing node analyzes the skin conductivity level signal and heart rate variability signal in real time to calculate the patient's real-time physiological stress index. A correlation model is established between the real-time physiological stress index and the intensity of rehabilitation training movements. During the training process, when the real-time physiological stress index continuously exceeds the individual adaptive threshold, the patient is determined to be in an overload state. Based on this determination, the local edge computing node automatically generates adjustment instructions to reduce training intensity, suggest pausing and resting, or switch to low-load actions; The adjustment instruction is sent to the interactive interface connected to the local edge computing node and displayed thereon. After the patient confirms the instruction, the patient is guided to execute the adjusted training plan.

[0013] As a further aspect of the present invention, an offline long-term rehabilitation progress assessment step is also included: The local edge computing node will locally encrypt and store all of the patient's standardized multimodal perception data frames, all generated event labels, and scoring data daily. At the designated long-term assessment nodes, the encrypted long-term data will be transmitted in batches to the regional rehabilitation monitoring center server. The regional rehabilitation monitoring center server decodes and analyzes the received long-term data to calculate the average rate of improvement in the patient's range of motion, the trend line of gait parameters converging towards normal values, and the decreasing curve of the frequency of abnormal events throughout the entire rehabilitation cycle. Based on the average rate of progress, trend line, and decline curve, a visualized long-term rehabilitation progress assessment report is generated for clinical review and rehabilitation plan review. The calculation of the patient's average rate of improvement in joint range of motion throughout the entire rehabilitation period, the trend line of gait parameters converging towards normal values, and the decreasing curve of the frequency of abnormal events specifically includes: From long-term data, the maximum effective range of motion is extracted from the daily range of motion angle change sequence of the specified target joint. A linear regression method is used to fit the curve of the maximum effective range of motion changing over time, and the slope of the curve is calculated as the average rate of progress. From long-term data, key gait parameters are extracted from the spatiotemporal parameter set of the gait cycle on a daily basis. The Euclidean distance between the daily key gait parameters and the standard normal value is calculated. The trajectory of the Euclidean distance changing over time is plotted, and a trend line of convergence of the gait parameters to the normal value is obtained by fitting. The number of attitude anomaly event tags recorded daily in long-term data is counted, the moving average value of the event over time is calculated, the time-series curve of the moving average value is plotted, and the decreasing curve of the frequency of the anomaly event is obtained by analysis. The average rate of progress, trend line, and decline curve are correlated with the patient's demographic and surgical information and integrated into a structured rehabilitation progress data record, which is used to generate the visualized long-term rehabilitation progress assessment report.

[0014] As a further aspect of the present invention, the present invention also includes an orthopedic rehabilitation monitoring network system based on edge intelligent decision-making technology. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology as described above.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A lightweight skeletal muscle biomechanical model is built into the local edge computing node, enabling real-time computation of standardized multimodal sensing data frames. This technology allows advanced biomechanical features such as joint angles, electromyographic envelopes, and gait parameters to be extracted instantly at the data acquisition source, eliminating the need to transmit massive amounts of raw data to remote locations. This significantly reduces the system's continuous network bandwidth consumption and minimizes power consumption and latency during data transmission. The feature extraction process is completed locally, avoiding the enormous computational overhead of processing raw data streams for the central server, allowing the system architecture to support the access and concurrent monitoring of more terminal nodes.

[0016] A pre-trained posture anomaly detection decision tree is deployed on local edge computing nodes to perform synchronous frame-by-frame comparison and analysis of the real-time calculated joint range of motion sequences, electromyographic envelope signals, and gait spatiotemporal parameter sets. This technology performs parallel logical decisions on three types of features based on preset safety thresholds, expected outbreak patterns, and standard templates. Once any feature exceeds its limit, an event alert is immediately generated locally with a specific anomaly label and associated data slices. This multi-dimensional, on-site decision-making mechanism achieves extremely low-latency conversion from data to alerts, enabling subsequent processes such as image retrieval to be triggered instantly upon an anomaly, significantly improving the real-time nature of rehabilitation monitoring and the timeliness of intervention. The alert information is processed, lightweight semantic data, rather than the raw data stream, improving the efficiency of wireless transmission and central storage. Attached Figure Description

[0017] Figure 1 This is a flowchart of the orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology described in this invention; Figure 2 A schematic diagram illustrating the working principle of anomaly detection and data slice generation for pre-trained pose anomaly detection decision trees. Figure 3 A graph showing the changes in abnormal events and physiological stress indices at each stage of rehabilitation; Figure 4 A graph showing the trend of joint range of motion and gait symmetry during rehabilitation; Figure 5 A chart showing the long-term trend of athletic performance scores. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 In the patient rehabilitation environment, standardized multimodal sensing data is first acquired. The local edge computing node then invokes its built-in lightweight skeletal muscle biomechanical model to perform real-time computation on the aggregated standardized multimodal sensing data frames. This computation process extracts three core biomechanical parameters: an angle change sequence reflecting joint range of motion, an electromyographic envelope signal reflecting muscle load, and a set of spatiotemporal parameters reflecting gait cycle. Subsequently, within the local edge computing node, these three types of parameters are input into a pre-trained posture anomaly detection decision tree for frame-by-frame comparison and analysis. This decision tree incorporates multiple judgment logics. When the angle change sequence of joint range of motion exceeds a safe threshold range, or the electromyographic envelope signal exhibits an unexpected burst pattern, or the spatiotemporal parameter set of the gait cycle deviates from the standard template, the system generates a posture anomaly event label and caches the associated sensor data slice. Finally, the lightweight alarm information containing the posture anomaly event label and associated sensor data slice is uploaded to the regional rehabilitation monitoring center server via a wireless body area network gateway. The regional rehabilitation monitoring center server is responsible for receiving and persistently storing lightweight alarm information from multiple local edge computing nodes in the network. It can also trigger a process to retrieve high-definition images from the video monitoring equipment of a specified patient based on the alarm information, so that medical staff can remotely review them.

[0021] In one embodiment of the invention, the patient rehabilitation area can be a standardized single-person rehabilitation ward or a home rehabilitation space. The multimodal data sensing network deployed within the patient rehabilitation area consists of a wearable motion capture device, an embedded mattress pressure sensor array, and indoor positioning anchors. Specifically, the wearable motion capture device is an inertial measurement unit (IMU) sensor node worn on the main segments of the patient's limbs. Each IMU sensor node includes a triaxial accelerometer, a triaxial gyroscope, and surface electromyography (EMG) electrodes. Specifically, the embedded mattress pressure sensor array is a high-density pressure sensor grid laid under the hospital bed or rehabilitation training mattress. The indoor positioning anchors are ultra-wideband positioning base stations deployed on the ceiling or walls of the rehabilitation area. Specifically, the wearable motion capture device collects raw inertial data and raw EMG signals generated by human movement at a sampling frequency of 100 Hz. The embedded mattress pressure sensor array collects raw matrix data of body pressure distribution at a sampling frequency of 30 Hz. The indoor positioning anchors receive radio frequency signals from the patient's worn tag at an update frequency of 10 Hz and calculate the raw position coordinate flow. Local edge computing nodes aggregate raw streaming data from wearable motion capture devices, embedded mattress pressure sensor arrays, and indoor positioning anchors in real time via wired controller LAN bus or wireless Bluetooth protocol.

[0022] In practical implementation, local edge computing nodes perform timestamp synchronization and packet verification on the aggregated raw streaming data to form standardized multimodal sensing data frames. The timestamp synchronization operation, based on the Network Time Protocol (NTP), assigns a uniform microsecond-level precision time stamp to each data packet from different sensing devices. The packet verification operation checks the frame header, frame tail, and checksum of each data packet, discarding packets that fail verification, and interpolating and completing the continuous data stream to form standardized multimodal sensing data frames with strict time alignment. In some embodiments, the local edge computing node invokes a built-in lightweight skeletal muscle biomechanics model to perform real-time computation on the standardized multimodal sensing data frames. From the standardized multimodal sensing data frames, inertial measurement data from wearable motion capture devices and a pressure distribution matrix from an embedded mattress pressure sensor array are separated. The inertial measurement data contains acceleration and angular velocity readings from multiple inertial measurement unit (IMU) sensor nodes, and the pressure distribution matrix is ​​a two-dimensional array containing M rows and N columns of pressure values.

[0023] In practical implementation, the human skeletal linkage model defined in the lightweight skeletal muscle biomechanical model and the forward kinematics algorithm are used to calculate the pose of inertial measurement data. The human skeletal linkage model simplifies the human body into a tree-like structure connected by fifteen rigid segments and fourteen joints. The forward kinematics algorithm starts from the root node, the pelvic segment, and recursively calculates the pose of each segment in the global coordinate system based on the joint rotation matrix between adjacent segments. The pose calculation outputs the real-time rotation angles and translations of the major joints in three-dimensional space, forming a sequence of angle changes in joint range of motion. The knee joint flexion angle change sequence in the sagittal plane is an example of the joint range of motion angle change sequence, containing a series of angle values ​​that change over time. In practical implementation, the pressure distribution matrix is ​​projected onto the center of gravity and the pressure center trajectory is tracked. Combined with the coordinates of the foot and hip joints in the human skeletal linkage model, the movement path of the plantar pressure center and the offset of the body pressure center in sitting and lying postures are calculated. The centroid projection calculation involves determining the weighted centroid coordinates of all pressure-sensing points in the pressure distribution matrix. The pressure center trajectory tracking calculation continuously calculates and records the coordinate position of the pressure center at each sampling moment. The foot pressure center movement path is the continuous movement trajectory of the pressure center point in the contact area between the foot and the ground from the heel to the toe during walking.

[0024] In practical implementation, the angular change sequence of joint range of motion is integrated with the movement path of the plantar pressure center. Through pre-defined gait phase segmentation logic within the model, continuous gait cycles are divided. The gait phase segmentation logic uses the first time the pressure value in the plantar pressure center movement path exceeds a set threshold as the starting point of the gait cycle, and the next time the pressure on the same side foot exceeds the set threshold as the ending point. Step length, step speed, and gait symmetry indices are extracted from each gait cycle to form a spatiotemporal parameter set for the gait cycle. The gait symmetry index is calculated by comparing the differences in step length or foot support phase time between consecutive left and right steps. The formula for calculating the gait symmetry index is as follows: in: Represents gait symmetry index, Represents the left step length. Represents the right step length. Gait symmetry index. The value ranges from zero to one, with a value closer to one indicating higher gait symmetry. In some embodiments, the raw electromyographic (EMG) signal channels in the inertial measurement data are filtered, rectified, and averaged to generate a smooth muscle activation level curve as the EMG envelope signal. The filtering process uses a bandpass filter of 20 Hz to 500 Hz to remove low-frequency motion artifacts and high-frequency noise from the raw EMG signal. The rectification process takes the absolute value of the filtered signal, and the moving average process smooths the rectified signal using a window with a time constant of 100 milliseconds. The EMG envelope signal generated by the rectus femoris muscle during a leg raise exhibits a single-peak shape, rising from the baseline to a peak and then falling back to the baseline; this single-peak shape is an example of the EMG envelope signal.

[0025] See Figure 2 In one embodiment of the present invention, the pre-trained posture anomaly detection decision tree contains multiple decision nodes defined by rehabilitation medicine experts. Each decision node corresponds to a specific safe threshold range for kinematic or physiological parameters and logical judgment rules. In specific implementation, the posture anomaly detection decision tree can be a binary tree structure data file stored in the memory of a local edge computing node. The root node of the decision tree corresponds to "action category determination," and the branch nodes correspond to specific rules such as "safe threshold for knee extension angle," "safe threshold for quadriceps electromyography root mean square," or "safe threshold for stride symmetry." In a scenario for postoperative rehabilitation training of the knee joint, the local edge computing node inputs the real-time calculated joint range of motion angle change sequence, electromyography envelope signal, and spatiotemporal parameter set of gait cycle into the posture anomaly detection decision tree at a rate of 100 frames per second for frame-by-frame comparison and analysis.

[0026] In practice, the flexion, extension, and internal / external rotation angles of the target joint at the current moment are extracted from the angular change sequence of joint range of motion. In active knee flexion and extension training, the postural anomaly detection decision tree sets the target joint as the affected knee. The flexion angle extracted from the angular change sequence of knee range of motion is the angle between the thigh and lower leg in the sagittal plane; the extension angle is the angle at which the knee returns from a flexed position to 0 degrees; and the internal / external rotation angle is the rotation angle of the lower leg around its longitudinal axis in the horizontal plane. In practice, the extracted flexion, extension, and internal / external rotation angles are compared with the corresponding safe threshold ranges for the joints in the postural anomaly detection decision tree. The safe threshold range for the second week of postoperative rehabilitation training for the knee joint in the postural anomaly detection decision tree can be set as follows: flexion angle 0 to 120 degrees, extension angle 0 degrees, and internal / external rotation angle -5 to 5 degrees. In practice, when the knee joint suddenly undergoes unexpected internal rotation during flexion, the extracted internal and external rotation angle is 10 degrees. The posture abnormality detection decision tree compares 10 degrees with the safe threshold range of -5 degrees to 5 degrees. 10 degrees exceeds the upper limit threshold of 5 degrees.

[0027] In practice, the root mean square (RMS) and peak values ​​of the electromyographic envelope signal are calculated within the current gait cycle or movement cycle. During the walking gait cycle, for the tibialis anterior muscle's EMS envelope signal on the affected side, the RMS value of the signal is calculated during the support phase from heel strike to toe liftoff, and the maximum value of the EMS envelope signal within this time period is identified as the peak value. In practice, the calculated RMS and peak values ​​are compared with the expected muscle activation patterns preset for the movement in the posture anomaly detection decision tree. The expected muscle activation pattern for flat-ground walking gait in the posture anomaly detection decision tree can be set to a RMS value of the tibialis anterior muscle's EMS envelope signal between 20 microvolts and 100 microvolts, with a peak value not exceeding 150 microvolts. In practice, if a patient overactivates the tibialis anterior muscle due to foot drop compensation, the calculated root mean square value is 180 microvolts, with a peak value reaching 250 microvolts. The posture abnormality detection decision tree compares the 180 microvolt root mean square value with the expected range of 20 to 100 microvolts; 180 microvolts exceeds the upper limit threshold of 100 microvolts. In some embodiments, the spatiotemporal parameter set of the gait cycle is used to calculate the difference between the current gait cycle and the standard rehabilitation stage template gait in terms of stride length and gait symmetry indices. Gait symmetry index difference. The calculation formula is expressed as follows: in: The degree of difference in gait symmetry index This represents the actual value of the gait symmetry index extracted from the spatiotemporal parameter set of the current gait cycle. This represents the preset template value of the gait symmetry index in the standard rehabilitation stage template gait. It can be understood that the gait symmetry index represents the degree of difference. The higher the value, the greater the deviation of the current gait from the standard template. In practice, if the actual value of the current gait symmetry index is... It is 0.7, while the standard template value is... The value was 0.9, and the difference in gait symmetry index was calculated. The value is 0.2. The gait symmetry index difference may be set in the posture anomaly detection decision tree. The threshold is 0.15, and 0.2 exceeds the threshold of 0.15.

[0028] In some embodiments, when any comparison result triggers the abnormal logic set in the posture anomaly detection decision tree, the system determines that the current frame data is abnormal. In the example of abnormal internal rotation of the knee joint, the comparison result of the internal and external rotation angles triggered the abnormal logic of "exceeding the limit of joint internal and external rotation angles". In the example of tibialis anterior muscle overactivation, the comparison result of the root mean square value of electromyography triggered the abnormal logic of "muscle overactivation". In specific implementation, the system packages the parameter type, actual value, and degree of exceeding the threshold range that trigger the anomaly to generate a posture anomaly event label. For abnormal internal rotation of the knee joint, the generated posture anomaly event label includes the parameter type "knee joint internal and external rotation angle", the actual value "10 degrees", the upper limit of the threshold "5 degrees", and the degree of exceeding "5 degrees". In specific implementation, the system simultaneously caches all sensor data within a set time window before and after triggering the anomaly, forming associated sensor data slices. Optionally, the time window can be set to 1.5 seconds before the abnormal trigger and 0.5 seconds after the abnormal trigger. The associated sensor data slice includes the raw data of all inertial measurement unit sensor nodes, pressure distribution matrix data, and all intermediate and final parameters obtained within this 2-second time window.

[0029] In one embodiment of the present invention, the patient performs unassisted active knee flexion and extension training. During a normal rehabilitation training cycle without triggering abnormal posture event tags, the local edge computing node calculates the motion smoothness score, motion completion score, and endurance performance score of the current training movement based on the knee joint range of motion angle change sequence and the spatiotemporal parameter set of the gait cycle. In specific implementation, calculating the motion smoothness score requires performing second-order difference calculation on the joint range of motion angle change sequence to obtain the joint angular acceleration sequence. Taking the knee flexion angle change sequence within a complete knee flexion and extension cycle as an example, the angle change sequence contains continuous sampling points from 0 degrees to 90 degrees and back to 0 degrees. Performing second-order difference calculation on the angle change sequence calculates the change in the rate of change of adjacent angles to obtain the joint angular acceleration sequence reflecting the acceleration change. In practice, the number of jitters exceeding a preset smoothing threshold in the joint angular acceleration sequence is counted. The motion smoothness score is then evaluated based on the number of jitters. The preset smoothing threshold can be set to an angle change of no more than 0.5 degrees within each sampling interval. The unit of the joint angular acceleration sequence is degrees per square second. If five sudden accelerations or decelerations exceeding the threshold are detected within one motion cycle, the number of jitters is recorded as 5, and the motion smoothness score is calculated. The calculation formula can be expressed as: in: Represents the smoothness score of motion. This represents the number of jitters obtained from statistics. This represents the score deducted for each shake; it can be understood as a motion smoothness score. The higher the value, the smoother and more fluid the movement.

[0030] In practice, calculating the movement completion score requires dynamically time-warping the sequence of angle changes in joint range of motion with the target angle sequence in the standard rehabilitation movement library. The minimum alignment distance between the sequences is calculated. The target angle sequence of the standard knee flexion movement includes an ideal angle value that increases uniformly from 0 degrees to 90 degrees and then uniformly decreases back to 0 degrees. The patient's actual knee range of motion may reach different angles at different speeds. The dynamic time warping algorithm uses the flexion time axis to find the optimal alignment path between the two sequences. In practice, the movement completion score is evaluated based on the minimum alignment distance, which is the sum of the absolute values ​​of the angle differences between corresponding points in the two sequences after warping. If the minimum alignment distance is 15 degrees, the movement completion score is... The calculation can be expressed as ,in The score represents the degree of completion of the action. This represents the calculated minimum alignment distance. This represents the deduction factor for each degree of deviation. In some embodiments, calculating the endurance performance score requires statistically analyzing the proportion of time during which stride length and gait symmetry indices in the spatiotemporal parameter set of the gait cycle remain within acceptable ranges within the current continuous training period. The acceptable range can be set by the therapist, for example, a stride length between 0.4 meters and 0.6 meters, and a gait symmetry index greater than 0.85. In specific implementations, the endurance performance score is evaluated based on the duration proportion. If, in a 10-minute walking training session, the total time for both stride length and gait symmetry indices to be within acceptable ranges is 8.5 minutes, then the duration proportion is 0.85, and the endurance performance score is calculated accordingly. It can be directly proportional to the duration, that is ,in Represents endurance performance rating. This represents the proportion of the duration.

[0031] In practice, the local edge computing node compares the calculated motion smoothness score, movement completion score, and endurance performance score with the patient's personal best historical score record and the preset target values ​​of the rehabilitation plan. The patient's personal best historical score record is stored in the non-volatile memory of the local edge computing node; for example, the historical best motion smoothness score is 90, the historical best movement completion score is 88, and the historical best endurance performance score is 92. The preset target values ​​of the rehabilitation plan are set by the rehabilitation therapist in the system; for example, the target for the motion smoothness score in the current training cycle is 80, the target for the movement completion score is 85, and the target for the endurance performance score is 80. In practice, if the currently calculated motion smoothness score is 85, the movement completion score is 82, and the endurance performance score is 90, the local edge computing node compares 85 with the historical best score of 90 and the target score of 80, compares 82 with the historical best score of 88 and the target score of 85, and compares 90 with the historical best score of 92 and the target score of 80. Optionally, the comparison logic can be set to trigger one type of feedback when the current score is higher than the historical best score, and another type of feedback when the current score is lower than the target score.

[0032] In practice, based on the comparison results, corresponding encouraging voice commands, standard movement technique reminder voice commands, or intensity adjustment suggestion voice commands are selected from a preset voice prompt library. The preset voice prompt library is stored in the file system of the local edge computing node and contains multiple pre-recorded audio files. For example, the encouraging voice command is "Well done, please keep it up," the standard movement technique reminder voice command is "Please pay attention to controlling the speed of knee flexion and avoid sudden force," and the intensity adjustment suggestion voice command is "The current completion quality is high; it is recommended to add another set of training." In some embodiments, if the current motion smoothness score of 85 points exceeds the historical best score of 90 points for the first time, but has already exceeded the target score of 80 points, the comparison result may trigger the selection of an encouraging voice command. If the current movement completion score of 82 points is lower than the target score of 85 points, the comparison result may trigger the selection of a standard movement technique reminder voice command targeting movement completion. If the current endurance performance score of 90 points is much higher than the target score of 80 points and close to the historical best score of 92 points, the comparison result may trigger the selection of an intensity adjustment suggestion voice command. Optionally, the selection of voice commands can be achieved through a decision matrix, the input of which is the difference between the current score, the historical best score, and the target score, and the output is the identifier of the voice command.

[0033] In practice, a local voice broadcasting device connected to a local edge computing node broadcasts selected voice commands to the patient. This device can be a wired or wireless speaker integrated within the rehabilitation area. The local edge computing node sends the audio data of the selected voice command to the local voice broadcasting device via an audio interface or wireless network protocol. The local voice broadcasting device receives the audio data, converts it into sound waves, and broadcasts it to guide the patient in adjusting current or subsequent training movements.

[0034] See Figure 3 This chart shows the changes in the number of abnormal events and the physiological stress index at different stages of orthopedic rehabilitation. As the rehabilitation progresses from the acute phase to the maintenance phase, both the number of abnormal events and the physiological stress index show a significant and continuous decrease, indicating that rehabilitation intervention effectively reduces risk and the patient's physiological burden. During the acute phase, both indicators are at their highest levels, reflecting the instability of the patient's physical condition in the early postoperative period, with the greatest risk of movement and physiological stress. The indicators decrease significantly from the subacute phase to the recovery phase, indicating that the patient's physical function gradually recovers and their adaptability to rehabilitation training increases. From the intensive phase to the maintenance phase, the indicators drop to a lower level, indicating that the patient has been able to stably complete rehabilitation tasks and has entered the consolidation and maintenance phase. The two curves show a high degree of consistency, indicating that the occurrence of abnormal events is closely related to the patient's physiological stress state.

[0035] In one embodiment of the present invention, a regional rehabilitation monitoring center server receives and stores lightweight alarm information from multiple local edge computing nodes corresponding to multiple patients, forming a rehabilitation abnormal event log database. The rehabilitation abnormal event log database can be a structured database table, where each record includes a timestamp, patient identifier, posture abnormality event tag type, specific parameters, a storage pointer to the associated sensor data slice, and patient rehabilitation stage metadata and surgery type metadata. In specific implementations, the system periodically performs data mining on the rehabilitation abnormal event log database, with the period set to daily or weekly. All recorded posture abnormality event tags and their detailed parameter information, patient rehabilitation stage metadata, and surgery type metadata are extracted from the rehabilitation abnormal event log database. The patient rehabilitation stage metadata can be the number of weeks post-surgery, and the surgery type metadata can be "anterior cruciate ligament reconstruction" or "total knee replacement surgery." Refer to Table 1 for a simplified rehabilitation abnormal event log database data table.

[0036] Table 1: Example Data Table of Rehabilitation Abnormal Event Log Library In practice, cluster analysis is used to group abnormal posture event labels, identifying high-frequency abnormal event categories in patients at specific rehabilitation stages or with specific surgical types. The cluster analysis method can be the K-means algorithm, using rehabilitation stage, surgical type, and abnormal event label type as feature vectors. Based on the example data in Table 1, the cluster analysis may identify a cluster whose main characteristics are "surgical type = anterior cruciate ligament reconstruction, rehabilitation stage = 2-3 weeks post-surgery, abnormal event label = knee inextension". In practice, for each high-frequency abnormal event category, the corresponding associated sensor data slices are retrospectively analyzed to extract motion pattern features within a set time period before the abnormality occurred. For the "knee inextension" abnormality, the associated sensor data slices of patient PT001 within 5 seconds before the occurrence of abnormal events ID1 and 3 are retrospectively analyzed. The motion pattern features extracted from the slices may include "average hip extension angular velocity of the affected leg at the end of the stance phase" and "proportion of the swing phase time of the unaffected leg".

[0037] In practice, the extracted movement pattern features are compared with corresponding features in normal rehabilitation data to construct a risk prediction feature vector capable of providing early warning of potential abnormalities. Normal rehabilitation data comes from training cycles within the same rehabilitation phase that have not triggered abnormal events; the normal range for the "average hip extension angular velocity" is calculated to be 60 to 80 degrees per second. In practice, if the "average hip extension angular velocity" extracted from pre-abnormal data is 45 degrees per second, below the lower limit of the normal range, this feature can be constructed as the risk prediction feature "low hip extension angular velocity on the affected side at the end of the stance phase." Risk Prediction Feature Vector It can be expressed as ,in Represents the risk prediction feature vector. arrive Representing n specific feature states, for example This represents a Boolean state indicating that "the hip extension angular velocity on the affected side is less than 50 degrees per second at the end of the stance phase." In practice, the risk prediction feature vector and its corresponding abnormal event category are used as new judgment rules or threshold adjustment criteria and integrated into the updated posture anomaly detection decision tree configuration file. For the example above, in the updated posture anomaly detection decision tree configuration file, a new branch judgment node may be added before the original "knee extension angle" direct judgment node, with the content "If the hip extension angular velocity on the affected side remains below 50 degrees per second at the end of the stance phase, a warning of potential underextension risk is issued."

[0038] In practical implementation, based on the statistical patterns and common characteristics discovered, the safety threshold range and logical judgment rules in the pre-trained posture anomaly detection decision tree are optimized and updated to generate an updated posture anomaly detection decision tree configuration file. For example, data mining revealed that a large number of patients undergoing anterior cruciate ligament reconstruction showed a "low gait symmetry" alarm in week 4, but the parameter values ​​were between 0.70 and 0.75, while the original threshold was 0.80. In practical implementation, in conjunction with the opinions of rehabilitation physicians, the lower limit of the safety threshold for the "gait symmetry index" for such patients at this stage can be dynamically adjusted from 0.80 to 0.70. In practical implementation, the updated posture anomaly detection decision tree configuration file is distributed to the local edge computing nodes corresponding to each patient, replacing the original pre-trained posture anomaly detection decision tree, realizing the networked iteration of monitoring rules. The distribution of the configuration file can be completed through a secure network communication protocol. After receiving the new configuration file, the local edge computing node loads it into memory and replaces the old decision tree model.

[0039] In practical implementation, a skin conductance sensor and a heart rate sensor integrated into a wearable motion capture device simultaneously collect the patient's skin conductivity level and heart rate variability (HRV) signals. The skin conductance sensor measures minute changes in skin conductivity at a frequency of 10 times per second, and the HRV signal is obtained by analyzing the sequence of changes in the R-wave interval on an electrocardiogram (ECG). In practice, a local edge computing node analyzes the skin conductivity level and HRV signals in real time to calculate the patient's real-time physiological stress index. The calculation formula can be expressed as: in: Represents real-time physiological stress index, The slope of the normalized skin conductivity level signal within a short time window. This represents the ratio of low-frequency power to high-frequency power in the heart rate variability signal. and It is a weighting coefficient used to balance the two effects. This can be understood as the real-time physiological stress index. The higher the value, the higher the patient's physiological stress level.

[0040] In practice, a correlation model is established between the real-time physiological stress index and the intensity of rehabilitation training movements. During training, when the real-time physiological stress index continuously exceeds the individual adaptive threshold, the patient is determined to be in an overload state. The correlation model can be a simple linear relationship, for example, the intensity of "high knee exercises" is assigned a value of 5, and the intensity of "slow walking" is assigned a value of 2. The individual adaptive threshold is obtained by adding a percentage offset to the baseline real-time physiological stress index value measured in the patient's resting state. For example, if the baseline value is 0.2 and the offset is set to 50%, then the individual adaptive threshold is 0.3. In practice, if the patient's real-time physiological stress index during "high knee exercises" exceeds the individual adaptive threshold, the patient is considered to be in an overload state. If a patient's score remains above 0.35 for 30 consecutive seconds, exceeding the individual adaptive threshold of 0.3, they are considered to be in an overload state. In practice, based on this determination, the local edge computing node automatically generates adjustment instructions to reduce training intensity, suggest pausing and resting, or switch to low-intensity exercises. These instructions might be, "High load detected; it is recommended to pause the current training and rest for 1 minute," or "It is recommended to switch to low-intensity marching in place." In practice, these instructions are sent to an interactive interface connected to the local edge computing node. This interface can be a touchscreen installed within the rehabilitation area, where the instructions appear as text and icons. After patient confirmation, the system guides the patient through the adjusted training plan. The patient can confirm receipt of the instructions via buttons on the touchscreen, and the system then controls the training process or updates voice prompts accordingly.

[0041] See Figure 4 This is a trend analysis chart of joint range of motion and gait symmetry during the rehabilitation process. Joint range of motion steadily improved, increasing from 32 degrees in the first week post-surgery to 85 degrees in the sixth week, showing an overall linear upward trend. A significant jump occurred particularly between weeks 5 and 6, indicating that the patient's joint function recovery entered an accelerated phase. This reflects the positive impact of rehabilitation training on joint range of motion, consistent with the general pattern of orthopedic postoperative rehabilitation. The gait symmetry index was abnormal, remaining at 0 throughout the entire 6-week rehabilitation cycle. Normally, gait symmetry should be between 0 and 1, with 0 representing complete asymmetry and 1 representing complete symmetry. The patient's joint function recovery trend is positive, indicating that the current rehabilitation training program is effective. After week 6, the training intensity and complexity can be gradually increased based on the improvement in joint range of motion.

[0042] In one embodiment of the present invention, the local edge computing node locally encrypts and stores all standardized multimodal sensing data frames, all generated event tags, and scoring data of the patient daily. The encryption storage employs an advanced encryption standard algorithm and is stored in the solid-state storage medium of the local edge computing node. All standardized multimodal sensing data frames include raw inertial measurement data from a wearable motion capture device, a pressure distribution matrix from an embedded mattress pressure sensor array, and the calculated angular change sequence of joint range of motion, electromyographic envelope signals, and a set of spatiotemporal parameters for gait cycles. All generated event tags include postural abnormality event tags, and the scoring data includes motion smoothness scores, movement completion scores, and endurance performance scores. At a designated long-term assessment node, such as after the patient completes an eight-week rehabilitation phase, the local edge computing node transmits the encrypted long-term data in batches to the regional rehabilitation monitoring center server. The transmission process is conducted in a wireless local area network environment using a secure file transfer protocol.

[0043] In practice, the regional rehabilitation monitoring center server decodes and analyzes the received long-term data, calculating the average rate of improvement in the patient's range of motion throughout the rehabilitation cycle, the trend line of gait parameters converging towards normal values, and the decreasing curve of the frequency of abnormal events. When calculating the average rate of improvement in the patient's range of motion throughout the rehabilitation cycle, the maximum effective range of motion is extracted from the daily angular change sequence of the target joint from the long-term data. Taking the daily maximum flexion angle of the right knee joint as an example, over an eight-week rehabilitation cycle, the maximum flexion angle value achieved by the right knee joint during active flexion-extension training is identified daily from the angular change sequence of the range of motion, resulting in a set of fifty-six daily maximum flexion angle values ​​sorted by time. In practice, a linear regression method is used to fit the curve of the maximum effective range of motion changing over time. The linear regression method aims to find a straight line that minimizes the sum of squared errors between this line and the set of fifty-six daily maximum flexion angle values. The slope of this line reflects the average rate of change in the range of motion over time. Average rate of improvement. The calculation formula can be expressed as: in: Represents the average rate of progress. Represents the total number of days. Representing the The number of days relative to the start date of rehabilitation. Representing the The maximum effective range of motion of the joint extracted each day. This can be understood as the average rate of progress. A positive value indicates that the overall range of motion of the joint is improving.

[0044] In some embodiments, when calculating the trend line of gait parameters converging towards normal values, key gait parameters are extracted from the spatiotemporal parameter set of the gait cycle on a daily basis from long-term data. These key gait parameters can be gait symmetry indices. The Euclidean distance between the daily key gait parameters and the standard normal value is calculated. The standard normal value can be set as the average value of the gait symmetry indices of healthy individuals, for example, 1.0. The absolute difference between the actual daily value of the gait symmetry indices and the standard normal value of 1.0 is the daily Euclidean distance. In specific implementations, the trajectory of the Euclidean distance over time is plotted, and a trend line of gait parameters converging towards normal values ​​is fitted. The fitting can use multinomial regression or moving average smoothing methods. The trend line can visually demonstrate how the degree of deviation of the patient's gait parameters from normal values ​​changes over time. When calculating the decreasing curve of the frequency of abnormal events, the number of posture abnormal event tags recorded daily in the long-term data is counted. The number of posture abnormal event tags represents the total number of alarms triggered daily. In practice, the moving average of the number of attitude anomaly event tags over time is calculated, with a seven-day window. The daily average of the number of attitude anomaly event tags over the past seven days is used as the moving average for the current day. A time-series curve of the moving average is plotted, and the decreasing curve of the anomaly event frequency is analyzed, demonstrating the long-term trend of the anomaly event frequency.

[0045] In practice, the average rate of progress, trend line, and decline curve are linked with the patient's demographic and surgical information to form a structured rehabilitation progress data record. The patient's demographic information includes age and gender, while the surgical information includes the surgical name and date. This structured rehabilitation progress data record can be a single database record containing the patient identifier, assessment period, and calculated average rate of progress. The data includes numerical values, polynomial coefficients of the gait symmetry convergence trend line, key point data of the abnormal event frequency decline curve, and associated demographic and surgical information text fields. In some embodiments, a visualized long-term rehabilitation progress assessment report is generated based on the average rate of progress, trend line, and decline curve. This visualized long-term rehabilitation progress assessment report can be an electronic document containing multiple charts, such as a line graph showing the daily maximum joint range of motion and its linear regression fit line, a curve graph showing the convergence trend of the Euclidean distance of the gait parameter, and a bar chart showing the average frequency of weekly abnormal events. Optionally, the report may also include brief textual analysis conclusions based on the trend data for clinical review and rehabilitation plan review.

[0046] See Figure 5This is a long-term trend analysis chart of the exercise quality score, showing the dynamic changes of three core training assessment indicators for patients from week 1 to week 8 after orthopedic rehabilitation surgery. All three indicators show a continuous increase from low to high scores, and the fluctuation range gradually decreases as the rehabilitation cycle progresses, reflecting the patient's increasing control over rehabilitation training and enhanced movement stability and consistency. From week 1 to week 3, all three indicators were in the low range with significant fluctuations, especially endurance performance, which repeatedly fell below 50 points, consistent with the clinical characteristics of weak muscle strength and poor movement control in the early postoperative period. From week 4 to week 6, movement smoothness and movement completion rapidly exceeded 90 points, and endurance performance recovered from its low point and steadily increased, marking the transition of rehabilitation training from the "adaptation" to the "intensification" stage. From week 7 to week 8, all three indicators remained in the high range with minimal fluctuations. Movement smoothness and movement completion tended to be consistent, and although endurance performance was slightly lower, it had stabilized, indicating that the patient's rehabilitation effect was significant and approaching the clinical rehabilitation goal.

[0047] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology, characterized in that, The method includes: Standardized multimodal sensing data is acquired, and the local edge computing node calls the built-in lightweight skeletal muscle biomechanical model to perform real-time calculation on the standardized multimodal sensing data frame, extracting the angle change sequence reflecting joint range of motion, the electromyographic envelope signal reflecting muscle load, and the set of spatiotemporal parameters reflecting gait cycle. In the local edge computing node, the angle change sequence of the joint range of motion, the electromyographic envelope signal, and the spatiotemporal parameter set of the gait cycle are input into a pre-trained posture anomaly detection decision tree for frame-by-frame comparison and analysis. When the angle change sequence of the joint range of motion exceeds the safe threshold range, or the electromyographic envelope signal shows an unexpected burst mode, or the spatiotemporal parameter set of the gait cycle deviates from the standard template, posture anomaly event labels and associated sensor data slices are generated. The lightweight alarm information, which includes the posture abnormality event tag and associated sensor data slices, is transmitted to the regional rehabilitation monitoring center server via a wireless body area network gateway. The regional rehabilitation monitoring center server is used to receive and persistently store the lightweight alarm information from multiple local edge computing nodes, and at the same time triggers a process to retrieve high-definition images from the video monitoring equipment of the specified patient.

2. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 1, characterized in that, The acquisition of standardized multimodal sensing data includes: A multimodal data sensing network is deployed in the patient rehabilitation area. The multimodal data sensing network consists of a wearable motion capture device, an embedded mattress pressure sensing array, and indoor positioning anchors. It is used to concurrently collect raw streaming data of the patient's skeletal muscle movement trajectory, body pressure distribution cloud map, and indoor activity path. Local edge computing nodes are deployed within the rehabilitation area to aggregate the raw streaming data collected by the wearable motion capture device, the embedded mattress pressure sensing array, and the indoor positioning anchor points to the local edge computing nodes in real time. The local edge computing nodes perform timestamp synchronization and data packet verification on the aggregated raw streaming data to form standardized multimodal sensing data frames.

3. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 2, characterized in that, The process of calling a built-in lightweight skeletal muscle biomechanics model from a local edge computing node to perform real-time computation on the standardized multimodal sensing data frame includes: From the standardized multimodal sensing data frame, inertial measurement data from the wearable motion capture device and pressure distribution matrix from the embedded mattress pressure sensing array are separated; Using the human skeletal linkage model and forward kinematics algorithm defined in the lightweight skeletal muscle biomechanical model, the pose calculation is performed on the inertial measurement data, and the real-time rotation angle and translation amount of the main human joints in three-dimensional space are output, forming the angular change sequence of the joint range of motion. The pressure distribution matrix is ​​subjected to centroid projection and pressure center trajectory tracking calculation. Combined with the coordinates of the foot and hip joints of the human skeletal linkage model, the movement path of the foot pressure center and the offset of the body pressure center in sitting and lying postures are calculated. By integrating the angular change sequence of the joint range of motion with the movement path of the plantar pressure center, and using the gait phase segmentation logic pre-set in the model, continuous gait cycles are divided, and stride length, stride speed, and gait symmetry indicators are extracted from each gait cycle to form a set of spatiotemporal parameters for the gait cycle. The raw electromyographic signal channels in the inertial measurement data are filtered, rectified, and subjected to moving average processing to generate a smooth muscle activation level curve, which serves as the electromyographic envelope signal.

4. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 3, characterized in that, The input is fed into a pre-trained pose anomaly detection decision tree for frame-by-frame comparison and analysis, including: The pre-trained posture anomaly detection decision tree contains multiple decision nodes defined by rehabilitation medicine experts. Each decision node corresponds to a specific safety threshold range of kinematic or physiological parameters and logical judgment rules. From the angular change sequence of the joint range of motion, extract the flexion angle, extension angle and internal / external rotation angle of the target joint at the current moment, and compare them with the safety threshold range of the corresponding joint in the decision tree; For the electromyographic envelope signal, calculate its root mean square value and peak value in the current gait cycle or movement cycle, and compare it with the expected muscle activation pattern preset for the movement in the decision tree. For the set of spatiotemporal parameters of the gait cycle, calculate the degree of difference between the current gait cycle and the standard rehabilitation stage template gait in terms of stride length and gait symmetry indicators; When any comparison result triggers the abnormal logic set in the decision tree, the current frame data is determined to be abnormal, and the parameter type, actual value, and degree of exceeding the threshold range that triggered the abnormality are packaged to generate the attitude abnormality event label. At the same time, all sensor data within the set time window before and after the abnormality is triggered are cached to form the associated sensor data slice.

5. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 4, characterized in that, It also includes real-time motion quality scoring and adaptive prompting steps performed at the local edge computing node: During a normal rehabilitation training cycle in which the abnormal posture event label is not triggered, the local edge computing node calculates the motion smoothness score, motion completion score, and endurance performance score of the current training action based on the angle change sequence of the joint range of motion and the spatiotemporal parameter set of the gait cycle. The calculated motion smoothness score, movement completion score, and endurance performance score are compared with the patient's personal best historical score record and the preset target value of the rehabilitation plan. Based on the comparison results, select the corresponding encouraging voice commands, standard action key reminder voice commands, or intensity adjustment suggestion voice commands from the preset voice prompt library; The local voice broadcasting device connected to the local edge computing node broadcasts the selected voice commands to the patient, guiding the patient to adjust the current or subsequent training actions.

6. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 5, characterized in that, The calculation of the motion smoothness score, motion completion score, and endurance performance score for the current training movement includes: The second-order difference calculation is performed on the joint range of motion angle change sequence to obtain the joint angular acceleration sequence. The number of jitters exceeding a preset smoothing threshold in the joint angular acceleration sequence is counted, and the motion smoothness score is evaluated based on the number of jitters. The joint range of motion angle change sequence is dynamically time-warped and matched with the target angle sequence in the standard rehabilitation movement library. The minimum alignment distance between the sequences is calculated, and the movement completion score is evaluated based on the minimum alignment distance. The endurance performance score is evaluated based on the proportion of time during which the stride length and gait symmetry index in the spatiotemporal parameter set of the gait cycle remain within the acceptable range during the current continuous training period.

7. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 6, characterized in that, It also includes in-depth analysis and dynamic adjustment steps of the rehabilitation plan performed on the server of the regional rehabilitation monitoring center: The regional rehabilitation monitoring center server receives and stores the lightweight alarm information from multiple local edge computing nodes corresponding to multiple patients, forming a rehabilitation abnormal event log library; Periodically perform data mining on the rehabilitation abnormal event log database to analyze the statistical patterns and common characteristics of the posture abnormal event tags in patient groups at different rehabilitation stages and with different surgical types; Based on the statistical patterns and common features discovered, the safety threshold range and logical judgment rules in the pre-trained posture anomaly detection decision tree are optimized and updated to generate an updated posture anomaly detection decision tree configuration file. The updated posture anomaly detection decision tree configuration file is distributed to the local edge computing node corresponding to each patient, replacing the original pre-trained posture anomaly detection decision tree, thereby realizing the networked iteration of monitoring rules. The periodic data mining of the rehabilitation abnormal event log database specifically includes: Extract all recorded postural abnormality event tags and their detailed parameter information, patient rehabilitation stage metadata, and surgical type metadata from the rehabilitation abnormality event log library; Cluster analysis was used to group the abnormal posture event labels and identify the categories of abnormal events that occur frequently in patients at specific rehabilitation stages or with specific surgical types. For each type of high-frequency abnormal event, backtrack and analyze the corresponding associated sensor data slices to extract motion pattern features within a set time period before the abnormality occurred. The extracted movement pattern features are compared with the corresponding features in normal rehabilitation data to construct a risk prediction feature vector that can provide early warning of potential abnormalities. The risk prediction feature vector and its corresponding abnormal event category are used as new judgment rules or threshold adjustment criteria and integrated into the updated attitude anomaly detection decision tree configuration file.

8. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 7, characterized in that, It also includes personalized load adjustment steps based on patient physiological feedback: By integrating a skin conductance sensor and a heart rate sensor into the wearable motion capture device, the patient's skin conductivity level signal and heart rate variability signal are collected simultaneously. The local edge computing node analyzes the skin conductivity level signal and heart rate variability signal in real time to calculate the patient's real-time physiological stress index. A correlation model is established between the real-time physiological stress index and the intensity of rehabilitation training movements. During the training process, when the real-time physiological stress index continuously exceeds the individual adaptive threshold, the patient is determined to be in an overload state. Based on this determination, the local edge computing node automatically generates adjustment instructions to reduce training intensity, suggest pausing and resting, or switch to low-load actions; The adjustment instruction is sent to the interactive interface connected to the local edge computing node and displayed thereon. After the patient confirms the instruction, the patient is guided to execute the adjusted training plan.

9. The orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology according to claim 8, characterized in that, It also includes offline long-term rehabilitation progress assessment steps: The local edge computing node will locally encrypt and store all of the patient's standardized multimodal perception data frames, all generated event labels, and scoring data daily. At the designated long-term assessment nodes, the encrypted long-term data will be transmitted in batches to the regional rehabilitation monitoring center server. The regional rehabilitation monitoring center server decodes and analyzes the received long-term data to calculate the average rate of improvement in the patient's range of motion, the trend line of gait parameters converging towards normal values, and the decreasing curve of the frequency of abnormal events throughout the entire rehabilitation cycle. Based on the average rate of progress, trend line, and decline curve, a visualized long-term rehabilitation progress assessment report is generated for clinical review and rehabilitation plan review. The calculation of the patient's average rate of improvement in joint range of motion throughout the entire rehabilitation period, the trend line of gait parameters converging towards normal values, and the decreasing curve of the frequency of abnormal events specifically includes: From long-term data, the maximum effective range of motion is extracted from the daily range of motion angle change sequence of the specified target joint. A linear regression method is used to fit the curve of the maximum effective range of motion changing over time, and the slope of the curve is calculated as the average rate of progress. From long-term data, key gait parameters are extracted from the spatiotemporal parameter set of the gait cycle on a daily basis. The Euclidean distance between the daily key gait parameters and the standard normal value is calculated. The trajectory of the Euclidean distance changing over time is plotted, and a trend line of convergence of the gait parameters to the normal value is obtained by fitting. The number of attitude anomaly event tags recorded daily in long-term data is counted, the moving average value of the event over time is calculated, the time-series curve of the moving average value is plotted, and the decreasing curve of the frequency of the anomaly event is obtained by analysis. The average rate of progress, trend line, and decline curve are correlated with the patient's demographic and surgical information and integrated into a structured rehabilitation progress data record, which is used to generate the visualized long-term rehabilitation progress assessment report.

10. An orthopedic rehabilitation monitoring network system based on edge intelligent decision-making technology, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the orthopedic rehabilitation monitoring network method based on edge intelligent decision-making technology as described in any one of claims 1 to 9.

Images