Ankle joint rehabilitation effectiveness evaluation method and system, evaluation terminal and storage medium

Abstract

The application provides a kind of ankle rehabilitation effectiveness evaluation method, system, evaluation terminal and storage medium, it is related to medical technical field.The weight allocation problem is modeled as a sequential decision model, the current dimension score and its trend are used as state input, and the cumulative improvement of rehabilitation effectiveness index in the future is used as the optimization goal.The strategy network trained by policy gradient algorithm can intelligently adjust the weight allocation according to the current rehabilitation situation of the patient, so that the weight of the three dimensions can be adaptively adjusted according to the current rehabilitation state and its trend of the patient, and the dynamic alignment of evaluation focus and actual weak link of the patient is realized.The rehabilitation effectiveness index is determined based on three dimensions, providing a multi-dimensional quantitative evaluation method for ankle rehabilitation, and finally generating intuitive rehabilitation effectiveness prompt information, realizing the objective quantification and visual feedback of ankle rehabilitation process.

Description

Technical Field

[0001] This application relates to the field of medical technology, and more specifically, to a method, system, assessment terminal, and storage medium for assessing the effectiveness of ankle joint rehabilitation. Background Technology

[0002] Ankle sprains are one of the most common types of sports injuries. Improper rehabilitation can easily lead to sequelae such as chronic instability, recurrent sprains, and traumatic arthritis. Traditional ankle rehabilitation assessments mainly rely on the doctor's clinical experience and physical examination, as well as the patient's subjective feedback, such as the degree of pain and range of motion. This lacks objective and quantifiable indicators of rehabilitation progress, resulting in delayed adjustments to the rehabilitation plan, a higher risk of secondary injury, and poor patient compliance. Summary of the Invention

[0003] The purpose of this application is to provide a method, system, assessment terminal, and storage medium for evaluating the effectiveness of ankle joint rehabilitation, so as to achieve objective quantification and visual feedback on the ankle joint rehabilitation process.

[0004] In a first aspect, embodiments of this application provide a method for assessing the effectiveness of ankle joint rehabilitation, the method comprising: Acquire pressure data, which includes static pressure data and / or dynamic pressure data. The pressure data is collected by pressure sensors installed in various areas of the insole, and each area corresponds to the medial forefoot, lateral forefoot, and center heel position of the patient's foot. Based on the pressure data, the patient's weight-bearing balance score, left-right weight-bearing symmetry score, and weight-bearing capacity score are determined under the target state. The weight-bearing balance score is determined based on the difference in pressure data collected by pressure sensors at the medial and lateral positions of the forefoot under the target state. The left-right weight-bearing symmetry score is determined based on the difference in rehabilitation efficacy index between the patient's left foot and right foot under the target state. The weight-bearing capacity score is determined based on the patient's current weight-bearing capacity and baseline weight-bearing capacity. The target state includes static and / or dynamic states. The weight-bearing balance score, the left-right weight-bearing symmetry score, and the weight-bearing capacity score are weighted and calculated to obtain the rehabilitation efficacy index. The weight of each score is determined by a dynamic weight allocation strategy. The dynamic weight allocation strategy includes: constructing the patient's rehabilitation process as a sequential decision model, using the scores and trends of weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity in the current assessment period as the state space, the weight combination corresponding to the three dimensions as the action space, and the cumulative increase in the rehabilitation efficacy index in a preset future time period as the cumulative reward; training the sequential decision model through a strategy gradient algorithm, and outputting the optimal weight combination in the current state to maximize the cumulative reward. Rehabilitation efficacy prompts are generated based on the rehabilitation efficacy index.

[0005] In the above implementation process, the weight allocation problem is modeled as a sequential decision-making model. The current scores and trends of each dimension are used as state inputs, and the optimization objective is to maximize the cumulative improvement of the rehabilitation efficacy index over a future period. The policy network trained using the policy gradient algorithm can intelligently adjust the weight allocation based on the patient's current rehabilitation status. This allows the weights of the three dimensions—weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity—to adaptively adjust according to the patient's current rehabilitation state and its trends, achieving dynamic alignment between the assessment focus and the patient's actual weaknesses. Simultaneously, using the cumulative improvement of the future rehabilitation efficacy index as the optimization objective means that weight allocation is no longer limited to fitting the current state but actively guides the patient towards the long-term optimal rehabilitation trajectory. This effectively solves the intervention decision-making problem during the rehabilitation plateau period and significantly improves the accuracy of individualized assessment and the foresight of rehabilitation training guidance. The rehabilitation efficacy index is determined by comprehensively considering three dimensions: weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity. This provides a multi-dimensional quantitative assessment method for ankle rehabilitation, ultimately generating intuitive rehabilitation efficacy prompts. This achieves objective quantification and visual feedback on the ankle rehabilitation process, providing patients with clear guidance on their rehabilitation progress. At the same time, it provides reliable data for doctors to develop and adjust rehabilitation plans and for patients to manage their own rehabilitation.

[0006] Secondly, embodiments of this application provide an ankle joint rehabilitation efficacy assessment system, the system comprising: The insole is equipped with pressure sensors in various areas, which correspond to the medial forefoot, lateral forefoot, and heel center of the patient's foot. The pressure sensors are used to collect pressure data. The insole is also equipped with posture sensors, which are used to collect the patient's posture data. An evaluation terminal, which is used to perform the method provided in the first aspect above.

[0007] In the above implementation process, by setting only three pressure sensors in three key areas of the insole—the medial forefoot, the lateral forefoot, and the center of the heel—and combining them with posture sensors to collect posture data, the hardware cost and system complexity are significantly reduced. The evaluation terminal performs a dynamic-static fusion rehabilitation efficacy assessment method based on pressure and posture data, thus providing a multi-dimensional quantitative assessment means for ankle rehabilitation. Ultimately, it generates intuitive rehabilitation efficacy prompts, achieving objective quantification and visual feedback on the ankle rehabilitation process. This provides patients with clear guidance on rehabilitation progress and provides reliable data for doctors to formulate and adjust rehabilitation plans and for patients to manage themselves.

[0008] Optionally, the insole is further provided with a data acquisition module, which communicates with each pressure sensor and the posture sensor; The data acquisition module is used to acquire data using an interrupt mechanism that employs timer interrupts and acceleration threshold interrupts. The data acquisition module is used to check whether static pressure data has been acquired within the current time window under the timed interrupt mechanism. If it has not been acquired and both the pressure data and the attitude data meet the static conditions, it wakes up each pressure sensor to acquire static pressure data and enters sleep mode after the acquisition is completed. The data acquisition module is used to wake up each pressure sensor to collect dynamic pressure data under the acceleration threshold interruption mechanism, record the pressure peak value and its average value during the gait cycle, and enter sleep mode after the collection is completed.

[0009] In the above implementation process, intelligent wake-up and data acquisition in both static and dynamic states are achieved by setting a data acquisition module in the insole and adopting a dual interrupt mechanism combining timed interrupt and acceleration threshold interrupt. The timed interrupt is responsible for periodically checking and collecting static pressure data, while the acceleration threshold interrupt responds to motion events in real time and collects dynamic pressure data. After the acquisition is completed, the module immediately returns to sleep mode. This on-demand wake-up strategy significantly reduces system power consumption, enabling the insole to maintain all-weather monitoring capabilities while achieving long-term continuous operation. In addition, the dual interrupt mechanism naturally separates static and dynamic data, providing a high-quality data source for the subsequent state-specific assessment in the rehabilitation efficacy index, and avoiding energy waste and data redundancy caused by continuous high-frequency sampling.

[0010] Optionally, the insole is further provided with a communication module that communicates with the evaluation terminal, and / or, the insole is further provided with a lithium battery power supply module that powers the various devices within the insole, and / or, the insole is further provided with an offline storage module that stores the collected pressure data and attitude data.

[0011] In the above implementation process, the communication module enables wireless data transmission between the insole and the assessment terminal, supporting remote rehabilitation monitoring; the lithium battery power supply module provides stable power to the various components of the insole, and combined with the low-power acquisition mechanism, it can ensure continuous power supply for a long time to meet the needs of long-term home rehabilitation; the offline storage module ensures that pressure and posture data are not lost when the communication connection is interrupted or the assessment terminal is offline, and is automatically retransmitted after the connection is restored, thereby significantly improving the data integrity and ease of use of the system.

[0012] Optionally, a rehabilitation mini-program is deployed in the assessment terminal; The rehabilitation mini-program is used to generate and display a pressure heatmap based on the acquired pressure data. And / or, the rehabilitation mini-program is also used to generate and display a trend chart based on the rehabilitation efficacy index; And / or, the rehabilitation mini-program is also used to generate and display rehabilitation training guidance suggestions based on the rehabilitation efficacy index; And / or, the rehabilitation mini-program is also used to display the doctor's rehabilitation assessment suggestions.

[0013] Optionally, the system also includes a cloud-based rehabilitation data platform for storing patients' historical rehabilitation data, allowing doctors to remotely view and issue rehabilitation assessment suggestions. The cloud-based rehabilitation data platform communicates with the assessment terminal. By introducing a cloud-based rehabilitation data platform into the system and enabling it to communicate with the assessment terminal, centralized storage and remote sharing of patients' historical rehabilitation data are achieved. This allows doctors to upload and view patients' rehabilitation data anytime, anywhere, and issue targeted rehabilitation assessment suggestions accordingly.

[0014] Thirdly, embodiments of this application provide an evaluation terminal, including a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the method provided in the first aspect above are performed.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the method provided in the first aspect above.

[0016] Fifthly, embodiments of this application provide a computer program product, including computer program instructions, which, when read and executed by a processor, perform the steps of the method provided in the first aspect above.

[0017] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of an ankle joint rehabilitation efficacy assessment system provided in an embodiment of this application; Figure 2 A flowchart illustrating an ankle joint rehabilitation efficacy assessment method provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of an assessment terminal for performing an ankle joint rehabilitation efficacy assessment method, provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0021] It should be noted that the terms "system" and "network" in the embodiments of this invention can be used interchangeably. "Multiple" refers to two or more; therefore, in the embodiments of this invention, "multiple" can also be understood as "at least two". "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / ", unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0022] It should also be noted that all actions involving the acquisition of signals, information, or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where the application is located, and with the authorization granted by the owner of the relevant device.

[0023] This application provides a method for assessing the rehabilitation efficacy of the ankle joint. This method uses pressure sensors installed in three key areas—the medial forefoot, the lateral forefoot, and the center of the heel—to collect static and dynamic pressure data. This allows for the differentiation of rehabilitation performance under different conditions. Based on three dimensions—weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity—a comprehensive rehabilitation efficacy index is determined. This provides a multi-dimensional quantitative assessment method for ankle joint rehabilitation, ultimately generating intuitive rehabilitation efficacy prompts. It achieves objective quantification and visual feedback on the ankle joint rehabilitation process, providing patients with clear guidance on their rehabilitation progress. Simultaneously, it provides reliable data for doctors to develop and adjust rehabilitation plans and for patients to manage their own rehabilitation.

[0024] To facilitate understanding of this solution, the systems involved in this solution will be introduced below.

[0025] Please refer to Figure 1 , Figure 1 This is a schematic diagram of the structure of an ankle joint rehabilitation efficacy assessment system 100 provided in an embodiment of this application. The system 100 includes an insole 110 and an assessment terminal 120.

[0026] The insole 110 is equipped with pressure sensors 112 in each area, and each area corresponds to the medial forefoot, lateral forefoot, and heel center of the patient's foot. The pressure sensors 112 are used to collect pressure data. The insole 110 is also equipped with posture sensors 114, which are used to collect the patient's posture data.

[0027] The assessment terminal 120, such as a smartphone, tablet, or dedicated medical terminal, is used to receive data sent by the insole 110 and execute subsequent rehabilitation efficacy assessment methods. For example, if the assessment terminal 120 is used to perform subsequent rehabilitation efficacy assessments based on pressure data, that is, to execute the ankle joint rehabilitation efficacy assessment method provided in subsequent embodiments, the specific implementation of which can be found in the detailed description of subsequent embodiments. The structural design of the insole 110 is described below.

[0028] In actual use, the insole 110 is worn on the patient's feet to collect foot pressure data and human posture data.

[0029] The insole 110 can employ a multi-layered composite structure. For example, the bottom layer could be a flexible printed circuit board (FPC), the middle layer an elastic cushioning material (such as ethylene-vinyl acetate (EVA) foam), and the top layer a skin-friendly and breathable fabric. Three flexible pressure sensors are embedded between the middle and top layers, corresponding to three key plantar areas: The medial region of the forefoot: This region corresponds to the area below the head of the first metatarsal head, specifically the bony prominence extending from the base of the big toe towards the medial side of the foot. This area bears part of the body's weight during walking and standing and is an important fulcrum for the medial longitudinal arch.

[0030] Lateral forefoot region: Corresponding to the area below the fifth metatarsal head, specifically the bony prominence on the lateral side of the base of the little toe. This region contributes to the formation of the lateral longitudinal arch, bearing relatively little weight but maintaining foot stability.

[0031] Heel area: Directly below the calcaneal tuberosity, which is the central weight-bearing point of the heel. The calcaneus is the largest tarsal bone in the foot, bearing approximately 60% of the body weight.

[0032] The layout of the aforementioned pressure sensor 112 is designed for key weight-bearing areas in ankle rehabilitation, and can accurately collect the pressure distribution on the inner and outer sides of the forefoot and the heel, reflecting the force balance of the ankle joint.

[0033] The three flexible pressure sensors mentioned above can be collectively referred to as pressure sensors 112, used to collect pressure data from various areas in real time. Each pressure sensor 112 can be a piezoresistive flexible thin-film sensor with a measurement range of 0-500 kPa, a response time of less than 50 milliseconds, a recovery time of less than 30 milliseconds, and the ability to withstand at least 100,000 pressure cycles. When the sole of the foot contacts the ground and applies pressure to the pressure sensor 112, the resistance value of the pressure sensor 112 changes, and the pressure value can be calculated by measuring the change in resistance.

[0034] In addition, an attitude sensor 114 is also provided in the insole 110. The attitude sensor 114 can be a low-power six-axis inertial measurement unit (IMU), such as the MPU6050 model, which integrates a three-axis accelerometer and a three-axis gyroscope. The accelerometer is used to measure acceleration signals in three orthogonal directions (X, Y, and Z axes); the gyroscope is used to measure angular velocity. The attitude sensor 114 can collect the patient's attitude data, including the composite acceleration magnitude, acceleration components of each axis, angular velocity, etc., to determine the patient's motion state (such as stationary, walking, running, foot lifting, etc.) and step count. In this embodiment, the acceleration range of the attitude sensor 114 can be configured to ±2g, ±4g, or ±8g, and the sampling rate is set to 100 Hz by default.

[0035] During design, the thickness of the insole 110 can be controlled to 3-5 mm, allowing it to be directly inserted into the ankle stabilizer or used in ordinary athletic shoes. The bottom of the insole 110 features an anti-slip texture to prevent slippage during wear. Other modules within the insole 110 (such as a data acquisition module, communication module, lithium battery power supply module, and offline storage module) are encapsulated in a small TPU (Thermoplastic Polyurethane) waterproof shell, connected to three pressure sensors 112 via flexible cables. This waterproof shell can be fixed to the arch side of the insole 110 or attached to the outside of the ankle stabilizer via Velcro, avoiding direct pressure.

[0036] In the above implementation process, by setting only three pressure sensors 112 in three key areas of the insole 110—the medial side of the forefoot, the lateral side of the forefoot, and the center of the heel—and combining them with posture sensors 114 to collect posture data, the hardware cost and system complexity 100 are significantly reduced. The evaluation terminal 120 performs a dynamic-static fusion rehabilitation efficacy evaluation method based on pressure and posture data, thereby providing a multi-dimensional quantitative evaluation means for ankle joint rehabilitation. Finally, it generates intuitive rehabilitation efficacy prompts, realizing objective quantification and visual feedback on the ankle joint rehabilitation process. This provides patients with clear guidance on rehabilitation progress and provides reliable data for doctors to formulate and adjust rehabilitation plans and for patients to manage themselves.

[0037] To achieve low-power, long-lasting operation of the insole 110, this embodiment incorporates a data acquisition module within the insole 110. This module communicates with each pressure sensor 112 and posture sensor 114. The data acquisition module is based on a low-power microcontroller unit (MCU) and communicates with the three foot area pressure sensors 112 and posture sensors 114 respectively. It employs a dual interrupt mechanism combining timed interrupts and acceleration threshold interrupts to intelligently wake up the system 100 for data acquisition. After the acquisition task is completed, the system 100 immediately returns to sleep mode, thereby controlling the average operating current below 150 microamps (µA) and supporting continuous use for more than 90 days.

[0038] The data acquisition module uses a timer interrupt and an acceleration threshold interrupt mechanism for data acquisition. The timer interrupt is automatically triggered by the MCU's internal timer at preset fixed time intervals (e.g., 2 seconds). This timer interrupt has a low priority and is used to periodically check whether static pressure data acquisition is needed. The acceleration threshold interrupt is generated by the comparator inside the attitude sensor 114. When the composite acceleration modulus (i.e., the square root of the sum of the squares of the three-axis accelerations) exceeds a preset motion trigger threshold, the attitude sensor 114 sends a hardware interrupt signal to the microcontroller. This interrupt has a high priority and can be interpreted as waking up the microcontroller to ensure no motion events are missed.

[0039] The data acquisition module is used to check whether static pressure data has been acquired within the current time window under the timed interrupt mechanism. If it has not been acquired and both the pressure data and the attitude data meet the static conditions, it wakes up each pressure sensor 112 to acquire static pressure data. After the acquisition is completed, it enters sleep mode.

[0040] In practice, the data acquisition module internally configures a timer within the MCU, causing the timer to generate an interrupt every 2 seconds. After system 100 power-on initialization, it enters low-power sleep mode by default. When the timer interrupt occurs, the MCU is woken up and performs the following operations: 1) Check if static pressure data has been collected within the current time window (e.g., 1 hour): The MCU reads the status flag bit in memory. At the beginning of each hour, this flag bit is cleared to zero. Once static pressure data has been successfully collected within the hour, the flag bit is set to 1. If the flag bit is already 1, the MCU skips the subsequent collection steps and returns directly to sleep mode.

[0041] 2) If no data is collected, the current data of the attitude sensor 114 and pressure sensor 112 are read: The MCU continuously reads the instantaneous values ​​of the three pressure sensors 112 and the triaxial acceleration composite magnitude, and records the sampling duration (e.g., 3 seconds) and sampling frequency (e.g., 10Hz).

[0042] 3) Determine if the static conditions are met: The static conditions refer to the patient being in a static standing state with stable pressure on the soles of their feet. Specific criteria include: all three pressure data points are greater than the minimum load threshold; the standard deviation of the triaxial acceleration composite modulus is less than a preset static threshold; and the rate of change of the pressure data during continuous sampling is less than a preset rate of change. Otherwise, it is considered dynamic. Calculate the average value and standard deviation of the pressure data collected by each pressure sensor 112 within the aforementioned sampling time (e.g., 3 seconds), as well as the standard deviation of the acceleration composite modulus. When the data meets the aforementioned static conditions, it is considered static.

[0043] 4) Collect and store static pressure data: If the judgment is successful, the MCU will use the average pressure collected by each pressure sensor 112 within 3 seconds as the static pressure data for this hour, mark it with a timestamp, and store it in the offline storage module. At the same time, the static pressure data collection flag will be set to 1.

[0044] 5) After the data acquisition is completed, the MCU turns off the continuous sampling of the pressure sensor 112 and the attitude sensor 114, and re-enters the low-power sleep mode to wait for the next timer interrupt or acceleration threshold interrupt.

[0045] The data acquisition module is used to wake up each pressure sensor 112 to collect dynamic pressure data under the acceleration threshold interruption mechanism, record the pressure peak value and its average value during the gait cycle, and enter sleep mode after the acquisition is completed.

[0046] For example, the data acquisition module is configured with an acceleration threshold interrupt function for the attitude sensor 114. The motion trigger threshold is set to 0.15g, meaning that when the composite acceleration modulus exceeds 0.15g, the IMU's internal comparator sends a high-priority interrupt signal to the MCU. The specific implementation steps are as follows: 1) Interrupt Trigger and Wake-up: When the patient suddenly begins to walk or run, the foot acceleration increases rapidly, and the composite acceleration modulus exceeds 0.15g, the IMU immediately generates an interrupt signal. The MCU is awakened from sleep mode and responds to the acceleration threshold interrupt.

[0047] 2) Handshake Confirmation and De-jitter: To prevent false triggers (such as momentary jerking when a patient lifts their foot to scratch an itch), the MCU first continuously collects acceleration data for 0.2 seconds after wake-up and calculates the synthetic modulus value. If the synthetic modulus value is consistently greater than 0.15g during this period, it is confirmed as a real motion event; otherwise, it is considered a momentary jerking, the current data collection is abandoned, and the system returns to sleep mode.

[0048] 3) Initiate dynamic pressure data acquisition: After confirming movement, the MCU simultaneously activates three pressure sensors 112 and posture sensor 114 to continuously acquire data at a sampling frequency of 20Hz. The duration is preset to 5-10 seconds (configurable). This duration should cover at least 3-5 complete gait cycles.

[0049] 4) Gait Cycle Detection and Peak Extraction: During the data acquisition process, the MCU detects the gait cycle in real time. The detection method can be the zero-crossing method of the acceleration Z-axis (vertical direction): when the ground reaction force reaches its maximum, a negative peak appears in the vertical acceleration, and the time interval between two adjacent negative peaks is one gait cycle. For each identified gait cycle, the MCU records the maximum pressure value (peak value) of the three pressure sensors 112 within that cycle and temporarily stores it in the MCU's internal buffer.

[0050] 5) Calculate the peak average: After data acquisition, the MCU summarizes all gait cycles (e.g., 5 cycles) collected. For each pressure sensor 112, the arithmetic mean of the peak values ​​across all cycles is calculated. For example, if the peak values ​​of the medial forefoot sensor in the 5 cycles are 210N, 220N, 215N, 225N, and 218N, then the peak average = (210 + 220 + 215 + 225 + 218) / 5 = 217.6N. The peak average values ​​for the lateral forefoot and heel are calculated similarly.

[0051] 6) Storing dynamic feature data: The MCU packages the peak average value, gait cycle count, and acquisition time of the three channels obtained in this dynamic acquisition and stores them in the offline storage module. If multiple dynamic movements occur within an hour (e.g., the patient walks multiple times), the dynamic feature data of each time is stored independently and then aggregated again at the top of the hour to form a representative value for that hour (the average value of the peak average value of all dynamic acquisitions, or the maximum peak value is used for abnormal monitoring).

[0052] 7) Return to sleep mode: After storage is complete, the MCU turns off all sensors and re-enters low-power sleep mode, waiting for the next interrupt to trigger.

[0053] Through the coordinated operation of the aforementioned timed interrupt and acceleration threshold interrupt, the data acquisition module only wakes up System 100 when needed, remaining in deep sleep mode the rest of the time, thus achieving ultra-long battery life. For example, the low-power MCU's operating modes include sleep mode, acquisition mode, and data transmission mode. Sleep mode current is less than 10 microamps, acquisition mode current is less than 5 milliamps, and data transmission mode current is less than 15 milliamps. The hourly acquisition time is approximately 30 seconds static plus 60 seconds dynamic. Assuming 5 triggers, the hourly data transmission time is approximately 5 seconds, with an average current of approximately 150 microamps. With a battery capacity of 480 milliamps, the theoretical battery life is 120 days, with a design target of 90 days, leaving a margin for error.

[0054] In the above implementation process, by setting a data acquisition module in the insole 110 and adopting a dual interrupt mechanism combining timed interrupt and acceleration threshold interrupt, intelligent wake-up and data acquisition in both static and dynamic states are achieved. The timed interrupt is responsible for periodically checking and acquiring static pressure data, while the acceleration threshold interrupt responds to motion events in real time and acquires dynamic pressure data. After the acquisition is completed, the module immediately returns to sleep mode. This on-demand wake-up strategy significantly reduces the system's power consumption, enabling the insole 110 to maintain all-weather monitoring capabilities while achieving long-term continuous battery life. In addition, the dual interrupt mechanism naturally separates static and dynamic data, providing a high-quality data source for subsequent state-specific assessments in the rehabilitation efficacy index, and avoiding energy waste and data redundancy caused by continuous high-frequency sampling.

[0055] To achieve wireless transmission and reliable local storage of rehabilitation data, and to ensure the long-term autonomous operation of system 100 in the absence of an external power source, this solution can also include a communication module, a lithium battery power supply module, and / or an offline storage module in the insole 110.

[0056] The communication module communicates with the assessment terminal 120. This module can employ a Bluetooth Low Energy (BLE) 5.0 protocol stack wireless communication module. BLE technology is designed for low-power, intermittent data transmission, and its broadcast interval, connection interval, and transmit power are configurable to balance power consumption and transmission distance. The communication module is responsible for wirelessly transmitting pressure and posture data to the assessment terminal 120 (e.g., the patient's smartphone). Posture data may also be optional, as it does not need to be transmitted to the assessment terminal 120. To improve transmission success rate, this module can employ a timed reporting strategy rather than real-time streaming.

[0057] The lithium battery power supply module powers all the components within the insole 110. This module is an energy supply unit centered on a rechargeable or disposable lithium primary battery, including the battery itself, a power management chip, a voltage conversion circuit, and a battery protection circuit. This module provides a stable voltage to all active components within the insole 110 (pressure sensor 112, attitude sensor 114, data acquisition module, offline storage module, and communication module). Simultaneously, the power management within the module supports multiple operating modes (sleep, standby, acquisition, and transmission), dynamically adjusting the output according to the load to reduce reactive power loss.

[0058] The offline storage module stores the acquired pressure and posture data. It can utilize non-volatile memory, ensuring data retention even during power outages. This module temporarily stores pressure and posture data acquired by the data acquisition module when the assessment terminal 120 is out of Bluetooth range, the connection is unstable, or the patient is not carrying the terminal. Storage management employs a circular buffer design, automatically jumping back to the beginning when the write pointer reaches the end of the storage space and overwriting the oldest data block, ensuring that the most recent time period (e.g., 48 hours) is always retained.

[0059] In the above implementation process, the communication module enables wireless data transmission between the insole 110 and the assessment terminal 120, supporting remote rehabilitation monitoring; the lithium battery power supply module provides stable power to the various components of the insole 110, and combined with the low-power acquisition mechanism, it can ensure continuous power supply for a long time to meet the needs of long-term home rehabilitation; the offline storage module ensures that pressure and posture data are not lost when the communication connection is interrupted or the assessment terminal 120 is offline, and is automatically retransmitted after the connection is restored, thereby significantly improving the data integrity and ease of use of the system 100.

[0060] Based on the above embodiments, a rehabilitation mini-program is deployed in the evaluation terminal 120. This mini-program receives pressure and posture data uploaded by the insole 110 via Bluetooth Low Energy protocol, and processes, visualizes, and provides rehabilitation guidance for the data. The rehabilitation mini-program can be developed using the WeChat mini-program framework, supports iOS and Android systems 100, and does not require the installation of a separate application, making it convenient for patients and doctors to use.

[0061] As the core application of the 120 assessment terminal, the rehabilitation mini-program provides patients with an intuitive and convenient rehabilitation management interface. The mini-program may include modules such as real-time data visualization, rehabilitation progress reports, abnormality alerts, and personalized rehabilitation guidance.

[0062] The real-time data visualization module can dynamically display the pressure distribution of the medial forefoot, lateral forefoot, and heel through a three-zone pressure heat map of the sole, using a gradient color from green to red (supporting switching between static standing and dynamic walking modes), and is supplemented by a pressure distribution ratio curve to show the changing trend of the pressure ratio of each zone over the past 24 hours, helping patients understand their own weight-bearing balance at any time.

[0063] For example, a rehabilitation mini-program can be used to generate and display stress heatmaps based on the obtained stress data.

[0064] A pressure thermogram is a visual representation of pressure levels in different areas of the foot, using variations in color intensity or hue. This solution uses a simplified thermogram of three areas on the sole of the foot (medial forefoot, lateral forefoot, and heel). Each area is filled with a color that gradually changes from green (low pressure) to red (high pressure), and the specific pressure value is displayed in real time. The pressure thermogram helps patients intuitively understand whether their weight-bearing distribution is balanced. Simultaneously, the pressure distribution percentage curve is presented as a line graph showing the hourly changes in the pressure percentage of each area over the past 24 hours. The horizontal axis represents time, and the vertical axis represents percentage. Patients can slide the graph to view the distribution balance at historical moments.

[0065] The rehabilitation progress report module can automatically generate structured reports by day, week, and month. For example, the daily report is pushed out at 22:00 every night, including the average rehabilitation efficacy index (referred to as REI), scores for three sub-dimensions (weight-bearing balance, left-right symmetry, and weight-bearing capacity), total steps, and effective wearing time, along with a brief comment; the weekly report adds the rate of change compared to the previous week and a REI trend chart, marking the rehabilitation target lines set by the doctor; the monthly report highlights rehabilitation milestones (such as the dates when the first 0.5 or 0.7 is reached). All reports can be exported as images or PDFs for easy sharing with doctors.

[0066] The anomaly warning module can monitor pressure data and derived indicators in real time. For example, it triggers an overload warning when the peak pressure during dynamic walking exceeds the doctor's set safety threshold; it pushes a weight imbalance warning when the weight balance score is below 0.6 for three consecutive days; and it reminds users of low compliance when the average daily effective wearing time is less than 8 hours for five consecutive days. Warning information is sent through application service notifications and mini-program pop-ups, and stored in the "Message Center" for historical review. The anomaly warning module may also include wearing reminders and charging reminders.

[0067] The rehabilitation mini-program is also used to generate and display trend charts based on the rehabilitation effectiveness index.

[0068] The rehabilitation efficacy index is a comprehensive quantitative indicator used to assess ankle joint rehabilitation. Its acquisition method will be detailed in the subsequent method embodiments and will not be elaborated upon here. Its value ranges from 0 to 1, with higher values ​​indicating better rehabilitation outcomes. The trend graph refers to the curve showing the change in the rehabilitation efficacy index over time. Trend analysis can identify whether rehabilitation progress is steadily improving, plateauing, or deteriorating, providing a basis for clinical decision-making.

[0069] After receiving each reported data, the rehabilitation mini-program calls the REI calculation function in the cloud or locally to obtain the REI value for that time. When generating trend data, each REI can be aggregated daily to form a daily average REI (the median of the REI for all valid hours within a day, excluding hours with no data during sleep). Similarly, the weekly REI is the median of the daily average REI for a week. Outliers (such as a sudden drop in REI to 0, which may indicate not wearing the device) are automatically filtered out. Then, a line graph can be drawn using the mini-program's native icon component, with the horizontal axis representing time and the vertical axis representing the REI value. This line graph can be displayed to the patient or sent to the doctor.

[0070] The rehabilitation mini-program is also used to generate and display rehabilitation training guidance suggestions based on the rehabilitation effectiveness index.

[0071] The rehabilitation training guidance and suggestion module can automatically generate personalized rehabilitation suggestions based on the current REI value and historical trends. These suggestions include training type (such as static standing, partial weight-bearing walking, balance training), training duration (minutes / day), maximum load (kilograms of force or body weight percentage), and precautions. The suggestions will be dynamically adjusted as the REI increases.

[0072] During the demonstration, a "Today's Suggestion" card can be set up in the mini-program. The card displays rehabilitation training guidance suggestions, with an encouraging message at the top (such as "You have persisted with rehabilitation for 12 days, keep it up!"), and 3-5 specific suggestions in a list below. Each suggestion has a "Done" button on the right, which patients can click to record their completion status. The completion data will be included in the rehabilitation statistics. At the same time, the mini-program also provides video or animated demonstrations (such as ankle pump exercises and heel raises), which can be viewed by clicking on the suggestion title.

[0073] The rehabilitation mini-program is also used to display doctors' rehabilitation assessment suggestions.

[0074] The doctor's rehabilitation assessment suggestions refer to the text suggestions manually entered by the registered doctor after viewing the patient's rehabilitation data through the back-end cloud platform, such as "increase daily walking time to 30 minutes" or "pay attention to weight-bearing on the inner side of the forefoot." The mini-program will display these suggestions synchronously, and can be accompanied by the doctor's signature and timestamp.

[0075] The rehabilitation mini-program calls an HTTPS interface each time it starts (or silently retrieves it in the background every 4 hours) to check if there are any new doctor suggestions for the patient. If so, a red dot is displayed at the top of the mini-program's homepage. Opening the "Doctor Suggestions" page displays all historical suggestions in reverse chronological order, each including: doctor's name, department, suggestion content, and publication date. Unread suggestions are highlighted.

[0076] Patients can click "Read" below each suggestion and enter simple feedback (such as "I have increased lateral weight-bearing training as advised"). The feedback will be sent back to the cloud for doctors to review later. If a doctor issues advice requiring urgent attention (such as "Reduce weight-bearing, otherwise there is a risk of further injury"), the mini-program will send service notifications via subscription messages to ensure patients are informed in a timely manner.

[0077] On the rehabilitation training guidance and suggestions page, the mini-program will distinguish between "intelligent assistant suggestions" (generated by an algorithm) and "doctor's special suggestions" (entered by a doctor). Doctor's suggestions have higher priority and will be pinned to the top and highlighted with a special background. If the doctor's suggestions conflict with the automatic suggestions, the mini-program will prompt "Please follow the doctor's suggestions".

[0078] In this way, the rehabilitation mini-program not only transforms the raw data collected by Insole110 into intuitive and visual information, but also provides interactive solutions from automatic assessment to personalized suggestions and doctor-patient interaction, significantly improving the intelligence level of ankle joint rehabilitation monitoring and the patient experience.

[0079] The rehabilitation mini-program module can also integrate a user module, which supports user login, viewing of personal information and rehabilitation plans, etc.

[0080] In addition, the rehabilitation mini-program integrates an incentive module to improve patient adherence through rehabilitation points, achievement badges, and community leaderboards. The entire mini-program interface uses a blue and white color scheme, which is simple and user-friendly, and realizes a human-computer interaction solution from data visualization, report generation, abnormal warnings to personalized guidance and doctor-patient interaction.

[0081] The clinical application process includes six stages: baseline acquisition, rehabilitation monitoring, data synchronization, physician evaluation, patient feedback, and discharge assessment. In the baseline acquisition stage, when patients first wear the smart insole 110, they complete static standing (10 seconds) and natural walking (5-10 meters) under the guidance of a physician. The system automatically collects the static average value of three pressure zones and the dynamic peak average value within the gait cycle, storing these as rehabilitation baseline data (corresponding to static baseline weight-bearing and dynamic baseline weight-bearing, respectively), serving as a benchmark for subsequent weight-bearing capacity calculations and rehabilitation effect comparisons. In the rehabilitation monitoring stage, patients perform training according to the established rehabilitation plan in their daily lives (such as static standing, partial weight-bearing walking, ankle pump exercises, etc.). The smart insole 110 continuously collects hourly pressure data and step counts through a dual-interruption mechanism, temporarily storing the data in an offline storage module. During the data synchronization phase, the insole 110 automatically uploads the hourly summary data (average static pressure, average dynamic pressure peak, steps, etc.) to the rehabilitation mini-program via Bluetooth BLE every hour on the hour. If the phone is not nearby or the Bluetooth connection fails, the data will be retained in the insole 110 and will attempt to upload again at the next hour on the hour. Circular storage ensures that data is not lost for at least 48 hours. During the doctor's assessment phase, doctors log in to the cloud-based rehabilitation data platform to view the patient's Rehabilitation Effectiveness Index (REI) trend chart, pressure heatmap, sub-indicator scores (weight-bearing balance, left-right symmetry, weight-bearing capacity), and warning records. Combining clinical experience, doctors judge whether the rehabilitation progress is ideal and, if necessary, adjust the rehabilitation plan online (such as modifying the weight-bearing safety threshold, increasing or decreasing training intensity) and issue text suggestions. During the patient feedback phase, patients can view their pressure heatmap, REI change curve, and automatically generated personalized rehabilitation guidance in real time within the mini-program, while receiving assessment suggestions from doctors. Patients can click "read" to confirm and provide text feedback on the implementation status (e.g., "I have reduced weight-bearing walking as advised"). During the discharge assessment phase, if the patient's rehabilitation efficacy index reaches the preset discharge criteria for a consecutive week (e.g., REI ≥ 0.85 and left-right weight-bearing symmetry score ≥ 0.9), the system automatically generates a discharge assessment report, indicating "Rehabilitation goals have been achieved, and the patient can return to normal life," and recommends periodic follow-up examinations. If the criteria are not met, monitoring continues and the patient is reminded to contact their doctor to adjust the treatment plan. The entire process achieves seamless integration from data collection, intelligent analysis, remote intervention to decision support, significantly improving the standardization of ankle rehabilitation and patient compliance.

[0082] In order to achieve multi-patient management, long-term data storage and remote doctor-patient interaction, in some embodiments, the above-mentioned system 100 also includes a cloud-based rehabilitation data platform for storing patients' historical rehabilitation data, so that doctors can remotely view and issue rehabilitation assessment suggestions. The cloud-based rehabilitation data platform communicates with the assessment terminal 120.

[0083] The cloud-based rehabilitation data platform can be a software system 100 running on a cloud server, which can communicate bidirectionally with the assessment terminal 120 (rehabilitation mini-program) through relevant communication protocols, while providing doctors with a dedicated web management terminal or a doctor's version mini-program.

[0084] The patient's historical rehabilitation data may include hourly stress data uploaded by the patient every day (static average, dynamic peak average, number of steps), rehabilitation efficacy index (REI) and its sub-indicators (weight-bearing balance score, left and right weight-bearing symmetry score, weight-bearing capacity score), stress heat map snapshot, rehabilitation weekly report automatically generated by the mini program, patient compliance record (daily wearing time, training check-in status), and suggestion texts issued by the doctor.

[0085] Each time a patient uses the rehabilitation mini-program, the mini-program will synchronize the data from the insole 110 via Bluetooth and then report the parsed hourly summary records (including pressure values, steps, status flags, etc.) as well as the locally calculated REI value and pressure heatmap configuration to the cloud-based rehabilitation data platform.

[0086] After receiving the data, the cloud-based rehabilitation data platform first verifies the patient's identity and the integrity of the data, then stores the data in a stress data table (partitioned by day) in a MySQL database. Simultaneously, to accelerate chart loading on the doctor's end, the cloud-based rehabilitation data platform asynchronously triggers a data aggregation task: aggregating hourly data by day into daily average REI, weekly average REI, and percentiles for each sub-indicator. The aggregation results are stored in a cache (Redis) or a summary table. For the automatically generated weekly rehabilitation report PDF for patients every Monday, the cloud-based rehabilitation data platform uses a reporting engine (such as JasperReports) to generate a PDF file containing trend curves and key indicator comparisons, uploads it to object storage, and simultaneously generates a download link on the mini-program.

[0087] After logging into the web management interface (requiring two-factor authentication), the doctor selects a patient from the patient list. The front-end requests a data overview of that patient from the back-end. The back-end queries the database for the daily REI, weight-bearing capacity score, balance score, left-right weight-bearing symmetry score for the past 30 days, and hourly stress heatmap data (stress value and steps) for the past 7 days. The doctor uses ECharts to create interactive graphs and supports time range filtering (weekly / monthly / custom). The doctor can also click on a specific day to view a list of stress values ​​for all hours of that day and view the stress heatmap changes in an animated format. In addition, the cloud-based rehabilitation data platform will collect and display the patient's warning events (such as the number of times of excessive weight-bearing and the number of days of low compliance) to help doctors quickly locate problems.

[0088] After viewing patient data on the web, doctors can enter text in the "Rehabilitation Suggestion" input box, such as: "The patient's forefoot shows significant insufficient weight-bearing on the medial side. It is recommended to perform 10 minutes of weight-shifting training daily, slowly shifting weight from the lateral to the medial side. Follow-up appointment next week." After the doctor submits, the backend stores the suggestion in the suggestion table of the database (including doctor ID, patient ID, suggestion content, timestamp, and whether it is urgent). Simultaneously, the cloud-based rehabilitation data platform sends a template message to the patient's WeChat via the rehabilitation mini-program's server-side interface (such as subscribeMessage), stating, "You have a new doctor's rehabilitation suggestion; please open the mini-program to view it." After the patient opens the mini-program, it retrieves the list of unread suggestions via an HTTPS GET request and displays them in reverse chronological order on the "Doctor's Suggestions" page. For suggestions marked "urgent," the mini-program will use a pop-up reminder and read the suggestion content aloud. After reading, the patient can click the "Read" button, and the mini-program will send a confirmation status back to the cloud-based rehabilitation data platform. The doctor will then see the suggestion read receipt. In addition, doctors can also modify the patient's safety threshold, such as adjusting the dynamic weight-bearing limit from 30% of body weight to 40% of body weight. The cloud-based rehabilitation data platform stores this parameter in the patient configuration table and sends it out the next time the mini-program is synchronized. The insole 110 then updates the local warning threshold based on the parameter forwarded by the mini-program (written to the insole 110 Flash via Bluetooth).

[0089] In the above implementation process, by introducing a cloud-based rehabilitation data platform into system 100 and communicating with assessment terminal 120, centralized storage and remote sharing of patients' historical rehabilitation data are realized, enabling doctors to upload and view patients' rehabilitation data anytime and anywhere, and issue targeted rehabilitation assessment suggestions accordingly.

[0090] Based on the above system embodiments, the corresponding ankle joint rehabilitation efficacy assessment method is described below, which is applied to the assessment terminal, such as... Figure 2 As shown, the method includes the following steps: Step S210: Obtain pressure data.

[0091] The pressure data includes static pressure data and / or dynamic pressure data. The pressure data is collected by pressure sensors installed in various areas of the insole, with each area corresponding to the medial forefoot, lateral forefoot, and heel center of the patient's foot.

[0092] The structural design of the insole can be referred to the relevant description in the aforementioned embodiments. The insole integrates three flexible pressure sensors, corresponding to three key areas on the patient's foot: the medial forefoot (below the first metatarsal bone), the lateral forefoot (below the fifth metatarsal bone), and the heel (directly below the calcaneus). The data acquisition module in the insole employs a dual interrupt mechanism (timed interrupt and acceleration threshold interrupt) to distinguish different patient states and collect two types of pressure data respectively. Static pressure data refers to the average pressure of different areas of the sole of the foot measured when the patient's body is stable and there is no significant change in acceleration. It represents the weight distribution in a static standing state. For example, when the patient is in a static standing state and the acceleration fluctuation is less than a preset threshold, the insole records a set of average pressure values ​​for 3-5 seconds every hour, including the pressure values ​​of the three areas.

[0093] Dynamic pressure data refers to the maximum pressure values ​​reached in different areas of the foot during each gait cycle of a patient's walking process. These values ​​are averaged over multiple cycles and represent the peak load during exercise. For example, when a patient's walking or running causes acceleration to exceed a motion threshold (e.g., 0.15g), the insole continuously collects data at a frequency of 20Hz for 5-10 seconds, extracting the peak pressure values ​​in three areas within each gait cycle, and calculating the average of the peak pressure values ​​across all gait cycles within that hour. This data reflects the patient's impact load and propulsion capacity during dynamic movement.

[0094] The aggregated data uploaded to the evaluation terminal by the data acquisition module may include: timestamps, static stress data, dynamic stress data, and the number of steps taken within the current time period. The number of steps can also be used as a separate indicator to measure the evaluation performance. After receiving the data, the evaluation terminal first parses and stores it to obtain the stress data required for this evaluation.

[0095] Step S220: Based on the stress data, determine the patient's load balance score, left-right load symmetry score, and load-bearing capacity score under the target state.

[0096] Depending on the needs, static stress data and / or dynamic stress data can be selected to calculate the rehabilitation efficacy index. For example, static stress data can be used to calculate the rehabilitation efficacy index under static conditions, and dynamic stress data can be used to calculate the rehabilitation efficacy index under dynamic conditions. The target state corresponds to the state of the selected data source. The target state includes static and / or dynamic states. The rehabilitation efficacy index under the two states can be evaluated comprehensively or separately.

[0097] The rehabilitation efficacy index is a comprehensive quantitative indicator, determined based on the patient's weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity. Weight-bearing balance refers to the consistency of pressure distribution on the inner and outer sides of the forefoot, reflecting whether the patient consciously distributes their weight evenly to avoid bias caused by pain or fear. Left-right weight-bearing symmetry refers to the similarity of rehabilitation progress between the two feet, reflecting whether the patient exhibits compensatory overuse of the healthy side. Weight-bearing capacity refers to the proportion of current total weight-bearing relative to the healthy baseline, visually indicating how much load the patient can withstand.

[0098] When calculating the rehabilitation efficacy index, the patient's weight-bearing balance score, left-right weight-bearing symmetry score, and weight-bearing capacity score under the target state can be determined based on pressure data. The weight-bearing balance score is determined by the difference in pressure data collected by pressure sensors at the medial and lateral positions of the forefoot under the target state. The left-right weight-bearing symmetry score is determined by the difference in the rehabilitation efficacy index of the patient's left foot and right foot under the previously obtained target state. The weight-bearing capacity score is determined by the patient's current weight and baseline weight. Then, the weight-bearing balance score, left-right weight-bearing symmetry score, and weight-bearing capacity score are weighted and calculated to obtain the rehabilitation efficacy index.

[0099] The weight-bearing balance score reflects the evenness of weight distribution between the medial and lateral sides of the forefoot. This score is calculated based on the difference in pressure values ​​collected by the medial and lateral forefoot pressure sensors under target conditions. The smaller the difference, the higher the score, indicating that the patient can evenly distribute their weight between the medial and lateral sides of the forefoot without bias.

[0100] The pressure data uploaded by the data acquisition module can also include static and dynamic pressure data collected by pressure sensors in various areas. When calculating the load balance score, the evaluation terminal can first extract the medial forefoot pressure value under the target state from the received pressure data. and forefoot lateral pressure value The load balance score R measures the degree of difference between the two, and is calculated using the following formula:

[0101] The formula outputs a value between 0 and 1, where 1 indicates that the pressure on the inside and outside is exactly the same, and 0 indicates that the pressure on one side is much greater than that on the other side.

[0102] The left-right weight-bearing symmetry score reflects the degree of symmetry in the rehabilitation progress of the patient's left and right feet. This score is calculated based on the difference between the rehabilitation efficacy index of the left foot and the right foot under previously obtained target conditions. The smaller the difference, the higher the score, indicating that the functional recovery of both feet is synchronized.

[0103] This scoring requires the patient to wear both insoles (left and right) simultaneously, with valid data available on both sides. The assessment terminal first needs to obtain the previously acquired unilateral rehabilitation efficacy index (REI) for the left and right feet under the same target conditions. The unilateral REI is calculated as follows: for the plantar surface of that side, only the weight-bearing balance score and weight-bearing capacity score are weighted (because the symmetry index is not defined within a single side), with weights set according to clinical needs; for example, 0.5 for each foot in the early stages of rehabilitation, and a bias towards weight-bearing capacity in the middle and later stages. If the unilateral index has not been pre-calculated, the system first calculates the weight-bearing balance score and weight-bearing capacity score for the left and right feet respectively based on the first two steps, and then synthesizes the unilateral REI according to the weights.

[0104] After obtaining the left and right unilateral REI, the left and right load symmetry score K is defined as the complement of the difference between the two:

[0105] in, This represents the rehabilitation efficacy index of the patient's left foot under the previously calculated target state. This represents the rehabilitation efficacy index of the patient's right foot under the previously calculated target state. This represents the maximum value between the rehabilitation efficacy index of the patient's left foot and the rehabilitation efficacy index of the patient's right foot.

[0106] The weight-bearing capacity score reflects the degree of recovery of a patient's current total plantar weight-bearing capacity relative to their healthy baseline (or rehabilitation baseline). This score is calculated based on the ratio of the current weight-bearing capacity (the sum of pressure values ​​in three zones) at the target state to a pre-stored baseline weight-bearing capacity. A ratio closer to or greater than 1 indicates a higher score and better recovery in weight-bearing capacity.

[0107] The assessment terminal extracts pressure values ​​from three regions (medial forefoot, lateral forefoot, and heel) under target conditions from the pressure data and calculates the current total load. This can be the average or sum of pressure values ​​from three zones, while simultaneously retrieving the patient's baseline load under the target condition from local or cloud-based data. Baseline load can be the average or sum of pressure data collected by pressure sensors in three zones under static or dynamic conditions when the patient initially wears the insole. The load-bearing capacity score Z is defined as the current load. Baseline load The ratio:

[0108] Step S230: The weighted average scores of weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity are calculated to obtain the rehabilitation efficacy index.

[0109] The weights of each score are determined through a dynamic weight allocation strategy, which includes: constructing a sequential decision model for the patient's rehabilitation process; using the scores and trends of load balance, left-right load symmetry, and load capacity within the current assessment period as the state space; using the weight combinations corresponding to the three dimensions as the action space; and using the cumulative improvement of the rehabilitation efficacy index within a preset future time period as the cumulative reward. The sequential decision model is trained using a strategy gradient algorithm to output the optimal weight combination in the current state to maximize the cumulative reward.

[0110] The sequential decision-making model is a dynamic system that makes decisions sequentially over time. Each assessment, such as once a day, constitutes a decision step. The weight allocation decision made by the system in the current step not only affects the calculation result of the rehabilitation efficacy index but also influences the patient's subsequent training behavior through the generated rehabilitation suggestions, thereby changing the scores of each dimension in future assessment cycles. The sequential decision-making model can be a partially observable reinforcement learning model, a reinforcement learning model, etc.

[0111] The state space is a mathematical set of input features of the policy network, used to describe the patient's current rehabilitation status. In this scheme, the state space contains six dimensions of feature values: the load balance score (R), the left-right load symmetry score (K), the load-bearing capacity score (Z) in the current assessment period, and the changing trends of these three scores relative to the previous several assessment periods.

[0112] Trends are typically characterized by first-order differences (current value minus the previous period's value) or linear regression slopes (the fitted slope of the most recent 3-5 periods). A positive trend indicates improvement in that dimension, a negative trend indicates regression, and a trend close to zero indicates a plateau. These six features together constitute the input vector of the policy network, enabling the network to simultaneously perceive information from both the current level and the momentum of improvement.

[0113] The action space is the set of output variables of the policy network, namely the weight combination (w_R, w_K, w_Z) corresponding to the three evaluation dimensions. This weight combination must satisfy two basic constraints: non-negativity constraint, that is, the weight of each dimension is greater than or equal to zero, ensuring that no dimension is completely ignored; and normalization constraint, that is, the sum of the three weights is strictly equal to 1, ensuring that the rehabilitation efficacy index is always within the standard range of 0-1, which is convenient for comparison with historical data and clinical thresholds.

[0114] Cumulative return is a quantitative indicator that measures the overall return brought by a certain weight allocation strategy over a period of time in the future, and it is also the ultimate goal of strategy gradient algorithm optimization. In this scheme, cumulative return is defined as the sum of the positive increments of the rehabilitation efficacy index relative to the current value over a predetermined number of evaluation periods in the future, starting from the current decision step. A discount factor (γ, usually taken as 0.9-0.99) is applied to the long-term increment to balance the short-term and long-term returns.

[0115] The discount factor is a coefficient ranging from 0 to 1, used to adjust the weighting of future returns in cumulative returns. The closer γ is to 1, the more the system values ​​long-term recovery effects (i.e., the recovery trend in the next few weeks), and the strategy network will tend to choose weight combinations that "may not significantly improve in the short term but have better long-term effects"; the closer γ is to 0, the more the system focuses on short-term returns, and the strategy will be more conservative, tending to quickly obtain immediate REI improvements.

[0116] In this scheme, the default value of γ is 0.95, which takes into account both the safety and controllability in the short term (preventing the risk of excessive burden caused by overly aggressive strategies) and fully incorporates the guiding role of long-term rehabilitation goals.

[0117] Policy gradient algorithms are a core method in reinforcement learning used to directly optimize the parameters of policy networks. The basic idea is to collect multiple complete decision trajectories through Monte Carlo sampling, calculate the cumulative reward for each decision step in the trajectory, and then update the network parameters in directions that increase the probability of "high-reward actions" while decreasing the probability of "low-reward actions," ultimately causing the output of the policy network to gradually converge to the optimal weight allocation policy.

[0118] The parameter update formula used in this scheme is: , where θ is the parameter of the policy network and α is the learning rate (e.g., 0.001). θ This represents the gradient of the parameter θ, π. θ (a t |s t ) indicates the current state s t Choose action a t The probability of (i.e., the weighted combination), G t This represents the cumulative reward for that decision step. The intuitive meaning of this formula is: if a weight combination brings a positive cumulative reward, the probability of that combination being selected again increases; conversely, it decreases the probability.

[0119] A policy network is a neural network model that maps from the state space to the action space. This scheme uses a two-layer fully connected feedforward network. The hidden layer contains 64 neurons with ReLU (Modified Linear Unit) activation. The output layer uses Softmax activation to ensure that the three weights are non-negative and sum to 1. The input layer has a dimension of 6 (corresponding to the 6 features in the state space), and the output layer has a dimension of 3 (corresponding to the weights in the three dimensions).

[0120] During the training phase, the network parameters are continuously updated using the policy gradient algorithm; during the inference phase, the network only performs forward propagation, directly outputting the weight combination after inputting the current state vector, which requires very little computation and can be completed in real time on the evaluation terminal.

[0121] The core of this implementation lies in transforming the weight allocation problem of the three dimensions of the rehabilitation efficacy index into a sequential decision optimization problem. By using the policy gradient algorithm in reinforcement learning, the weights are dynamically and adaptively adjusted, thus enabling the rehabilitation efficacy index to not only reflect the current state but also guide patients towards their long-term optimal rehabilitation trajectory. The specific implementation process is as follows: First, a reinforcement learning-based environmental interaction framework is constructed. Each rehabilitation assessment cycle (e.g., daily or every three days) is defined as a decision step. At each decision step, the system acquires the weight-bearing balance score (R), left-right weight-bearing symmetry score (K), and weight-bearing capacity score (Z) calculated within the current assessment cycle, and simultaneously calculates the trends of these three scores relative to the previous few assessment cycles. These trends can be represented using first-order differences (i.e., the current value minus the previous cycle's value) or linear regression slopes (taking the fitted slope of the most recent 3-5 cycles) to reflect the rate of improvement or deterioration trend in each dimension. The system uses these six values—the R, K, and Z scores for the current cycle and their respective trends—to form the state space, which serves as the input feature vector for the policy network. For example, if a patient's current cycle R=0.65 (an increase of 0.05 from last week), K=0.70 (the same as last week), and Z=0.50 (an increase of 0.10 from last week), then the state vector can be represented as [0.65, 0.05, 0.70, 0.00, 0.50, 0.10].

[0122] Secondly, define the action space and policy network. The action space is a combination of weights (w_R, w_K, w_Z) corresponding to three dimensions, which must satisfy the non-negativity constraint (w_i≥0) and the normalization constraint (w_R+w_K+w_Z=1). The policy network adopts a parameterized neural network (e.g., a two-layer fully connected network with 64 hidden neurons and a softmax activation function in the output layer to ensure that the weights sum to 1). This network takes the current state vector as input and directly outputs a three-dimensional weight vector, i.e., π. θ(a|s), where θ is the network parameter, a is the output weight action, and s is the current state. In the early stage of training, the network parameters are randomly initialized, and the output weights are approximately uniformly distributed (approximately 0.33, 0.33, 0.34). As training progresses, the network gradually converges to the optimal policy.

[0123] Next, a cumulative return function is designed. Cumulative return is a core indicator for measuring long-term rehabilitation effectiveness and is also the target of strategy gradient algorithm optimization. In this implementation, the cumulative increase in rehabilitation efficacy index within a preset future time period (e.g., the next two weeks or the next five evaluation cycles) is defined as the cumulative return. The specific calculation method is as follows: starting from the current decision step, multiple decision steps are executed consecutively according to the current strategy (each step outputs new weights based on the new state and calculates the corresponding rehabilitation efficacy index). The sum of the positive increments of the rehabilitation efficacy index of these future steps relative to the current rehabilitation efficacy index is accumulated, and a discount factor γ is introduced to balance short-term and long-term returns. If the rehabilitation efficacy index of a future step is lower than the current value, the contribution of that step is a negative return, thus penalizing the weight decisions that lead to rehabilitation regression. The mathematical expression of the cumulative return is: REI t REI is the rehabilitation efficacy index for the current cycle. t+k Let REI be the rehabilitation efficacy index for the k-th future period, and T be the preset number of future steps (e.g., T=5). For example, if the current REI is 0.55, and a certain weighting strategy results in REIs of 0.58, 0.62, 0.67, 0.70, and 0.72 for the next 5 periods, the cumulative return will be approximately 0.4698. A positive value indicates that the strategy contributes to long-term improvement; if a strategy causes the future REI to decrease, the cumulative return may be negative.

[0124] Subsequently, a policy gradient algorithm (such as the REINFORCE algorithm or the Actor-Critic architecture) is used to train the policy network. The training phase can be conducted offline in the cloud, utilizing a large amount of anonymized historical rehabilitation data (including R, K, and Z sequences of different patients and their final rehabilitation outcomes) to simulate environmental interactions. Multiple complete decision trajectories are collected through Monte Carlo sampling, and the cumulative reward for each decision step in each trajectory is calculated, serving as the basis for weight update direction. The core idea of ​​policy gradient is to increase the probability of actions that bring positive cumulative rewards and decrease the probability of actions that bring negative cumulative rewards; the parameter update formula is as described above. Through several rounds of iterative training, the policy network gradually learns to output optimal weights under different state combinations. For example, when the state shows a low and declining Z score while R and K are high and stable, the network will output a high w_Z (e.g., 0.55) while decreasing w_R and w_K (approximately 0.225 each) to guide the assessment focus towards the weakest weight-bearing capacity dimension, prompting patients to strengthen weight-bearing training; when the state shows an improving trend in R and K while Z is close to baseline, the network will appropriately balance the weights to make the assessment more comprehensive.

[0125] After training, the policy network is deployed to the evaluation terminal for online inference. In actual evaluation, the evaluation terminal calculates R, K, Z and their trends for the current period, and inputs them into the trained policy network. The network then outputs the optimal weight combination for the current state through forward propagation. For example, for a mid-stage rehabilitation patient, the current state is R=0.72 (trend +0.02), K=0.65 (trend -0.03), Z=0.58 (trend +0.08). The policy network might output w_R=0.30, w_K=0.35, w_Z=0.35, indicating a slight regression in symmetry. The system automatically increases the weights of the symmetry dimension to strengthen monitoring.

[0126] Subsequently, the three weights are summed with the corresponding dimension scores, i.e., REI=w_R×R+w_K×K+w_Z×Z, to obtain the rehabilitation efficacy index for this study.

[0127] Finally, to ensure the long-term effectiveness of the strategy network, the system also incorporates an online fine-tuning mechanism. Every preset period (e.g., every two weeks), the system uses the patient's latest accumulated rehabilitation data to fine-tune the strategy network with a small number of gradient steps, gradually transitioning the weight allocation from group-wide commonality to individual specificity. Simultaneously, during inference, if the weight combination output by the strategy network conflicts with clinical safety rules (e.g., when K is below 0.3, the algorithm's output w_K is too low), the system activates a safety constraint layer to forcibly correct the weights (e.g., clamping w_K to no less than 0.30), ensuring that the weight allocation never violates basic safety monitoring requirements.

[0128] Step S240: Generate rehabilitation efficacy prompt information based on the rehabilitation efficacy index.

[0129] After obtaining the current Rehabilitation Effectiveness Index (REI), the assessment terminal can compare it with preset rehabilitation threshold stages to evaluate the rehabilitation progress and generate corresponding rehabilitation effectiveness prompts. For example, in the early stage of rehabilitation, static stress data is mainly used to calculate the rehabilitation effectiveness index. If the REI is less than 0.4, it indicates poor weight-bearing capacity, and rehabilitation effectiveness prompts can be generated, such as suggesting non-weight-bearing training as the main focus. In the middle stage of rehabilitation, both static and dynamic rehabilitation effectiveness indices can be used for evaluation. At this time, the dynamic rehabilitation effectiveness index may be lower than the static rehabilitation effectiveness index. It is necessary to monitor whether the dynamic rehabilitation effectiveness index is steadily improving and whether the static rehabilitation effectiveness index is maintained. If the static rehabilitation effectiveness index does not increase for a long period of time, walking function recovery is assessed as impaired, and rehabilitation effectiveness prompts can be given, such as being able to perform partial weight-bearing walking. In the later stage of rehabilitation, the dynamic rehabilitation effectiveness index can be used for evaluation. If the REI is greater than or equal to 0.7, it indicates the later stage of rehabilitation, and rehabilitation effectiveness prompts can be given, such as being able to attempt normal walking and paying attention to symmetry.

[0130] Additionally, the assessment terminal can compare this week's average REI with last week's. If the REI increase is less than 0.05 for two consecutive weeks and the current REI is still below 0.7, a rehabilitation efficacy warning message will be generated: rehabilitation progress is slow, and it is recommended to contact the doctor to adjust the plan; if the REI drops by more than 0.3, an early warning will be triggered.

[0131] In the above implementation process, the weight allocation problem is modeled as a sequential decision-making model. The current scores and trends of each dimension are used as state inputs, and the optimization objective is to maximize the cumulative improvement of the rehabilitation efficacy index over a future period. The policy network trained using the policy gradient algorithm can intelligently adjust the weight allocation based on the patient's current rehabilitation status. This allows the weights of the three dimensions—weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity—to adaptively adjust according to the patient's current rehabilitation state and its trends, achieving dynamic alignment between the assessment focus and the patient's actual weaknesses. Simultaneously, using the cumulative improvement of the future rehabilitation efficacy index as the optimization objective means that weight allocation is no longer limited to fitting the current state but actively guides the patient towards the long-term optimal rehabilitation trajectory. This effectively solves the intervention decision-making problem during the rehabilitation plateau period and significantly improves the accuracy of individualized assessment and the foresight of rehabilitation training guidance. The rehabilitation efficacy index is determined by comprehensively considering three dimensions: weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity. This provides a multi-dimensional quantitative assessment method for ankle rehabilitation, ultimately generating intuitive rehabilitation efficacy prompts. This achieves objective quantification and visual feedback on the ankle rehabilitation process, providing patients with clear guidance on their rehabilitation progress. At the same time, it provides reliable data for doctors to develop and adjust rehabilitation plans and for patients to manage their own rehabilitation.

[0132] Based on the above embodiments, the aforementioned left-right weight-bearing symmetry score is determined according to the degree of difference between the rehabilitation efficacy index of the patient's left foot and the rehabilitation efficacy index of the patient's right foot under the previously obtained target state. Addressing the common situation where patients only wear a single smart insole during the actual rehabilitation phase, this solution provides a method for estimating unilateral plantar pressure data based on biomechanical coupling, indirectly obtaining pressure data from the contralateral foot, thereby ensuring the calculability of the left-right weight-bearing symmetry score and the complete rehabilitation efficacy index. Its core basis lies in the fact that during the normal human gait cycle, there is a strict temporal symmetry and mechanical complementarity between the ground contact duration, swing phase duration, and plantar pressure rise pattern of both feet. The pressure distribution characteristics of one foot implicitly contain rich information about the corresponding state of the contralateral foot. In this situation, the rehabilitation efficacy index of the patient's left foot and the rehabilitation efficacy index of the patient's right foot can be obtained in the following way: when only the pressure data of the patient's unilateral foot is obtained, the gait timing parameters and pressure dynamic parameters under the target state are extracted from the pressure data; based on the gait timing parameters and pressure dynamic parameters, the estimated pressure data of the patient's contralateral foot under the corresponding target state is estimated using a preset mapping model; the first rehabilitation efficacy index is determined based on the pressure data of the unilateral foot, and the second rehabilitation efficacy index is determined based on the estimated pressure data of the contralateral foot.

[0133] Among them, gait timing parameters refer to the time metrics of each key phase of a single foot within a complete gait cycle, used to describe the time allocation pattern of the foot during walking. This scheme mainly includes two core metrics: ground contact duration and swing phase duration.

[0134] Ground contact duration refers to the time interval from the moment the heel strikes the ground to the moment the toes on the same side lift off the ground, reflecting the duration of weight load borne by that side's foot. Swing phase duration refers to the time interval from the moment the toes lift off the ground to the moment the heel on the same side strikes the ground again, reflecting the duration of the foot in a suspended swinging state. The sum of these two phases constitutes the duration of a complete gait cycle.

[0135] Dynamic pressure parameters refer to the dynamic characteristics of pressure changes over time in different areas of the foot during the gait cycle, used to characterize the force generation pattern and propulsion efficiency during foot strike and takeoff. This scheme mainly includes two indicators: the pressure rise slope and the time of peak pressure occurrence.

[0136] The pressure rise slope refers to the average rate of change of pressure over time from the moment the foot touches the ground until the pressure reaches its maximum value. It is typically calculated by linearly fitting the pressure-time curve of this rise segment using the least squares method. The slope of the fitted line is the pressure rise slope, measured in Newtons per second (N / s). This indicator reflects the speed and explosive force of the load-bearing capacity of that area; a larger rise slope indicates a faster load-bearing response and stronger propulsive force in that area.

[0137] Peak pressure occurs when the timer starts counting from the moment the heel strikes the ground until the pressure in that area reaches its maximum value. It is usually expressed as a percentage of the duration of support on that side or as an absolute time (milliseconds). This indicator reflects the timing of force exertion in that area during the gait cycle. For example, the heel area typically peaks in the early phase of the stance phase (approximately the first 20%), while the forefoot area typically peaks in the late phase of the stance phase (approximately 70%-80%), propelling the body forward.

[0138] The mapping model is a mathematical transformation relationship between gait parameters of one foot and pressure values ​​of the contralateral foot region. It is used to supplement bipedal data by estimating contralateral pressure when only unilateral measured data is available. In this scheme, the mapping model is constructed using a multiple regression method. Specifically, it uses gait temporal parameters and pressure dynamic parameters of one foot as input features, and the estimated pressure values ​​of three regions of the contralateral foot (medial forefoot, lateral forefoot, and center of the heel) as output targets. The mapping function is established by solving the regression coefficient matrix using the least squares method. Its mathematical model can be expressed as:

[0139] Among them, P contra,jx is the estimated pressure value for the j-th region of the contralateral foot. i For the i-th input feature (a total of 8, including support duration, swing phase duration, pressure rise slope and peak occurrence time in each of the three regions), β ij For the corresponding regression coefficient, β 0j For the intercept term, ε j This represents the random error term. The model training requires a large amount of bipedal synchronously collected sample data, and the least squares method is used to minimize the sum of squared residuals between the predicted and measured values.

[0140] Specifically, when the assessment terminal confirms that only effective pressure data of one foot (usually the affected side) of the patient has been acquired, the gait timing parameters and pressure dynamic parameters under the target state are first extracted from the pressure data. The gait timing parameters include the ground contact support duration and swing phase duration of the foot on one or more complete gait cycles. For example, the time interval from the moment the heel strikes the ground to the moment the toe leaves the ground is the support duration, and the time interval from the moment the toe leaves the ground to the moment the heel strikes the ground again on the same side is the swing phase duration. These two duration parameters reflect the weight-bearing time and airborne time of the foot on that side, and they have a strict inverse relationship with the contralateral foot in terms of time. The dynamic parameters of pressure include the pressure rise slope and the peak pressure occurrence time of the medial forefoot region, lateral forefoot region, and heel center region of the foot on that side during the gait cycle. The pressure rise slope refers to the rate of change of pressure value over time from the moment the foot touches the ground until the pressure reaches its peak value. It is usually obtained by fitting the linear slope of the rise segment using the least squares method. The peak pressure occurrence time refers to the time interval between the moment the heel touches the ground and the moment when the pressure in that region reaches its maximum value. This parameter reflects the force exertion sequence and propulsion efficiency of each region of the foot.

[0141] Subsequently, the extracted gait timing parameters and pressure dynamic parameters are used as input features and fed into a pre-defined mapping model to estimate the pressure data of each region of the contralateral foot under the corresponding target state. This mapping model is a mathematical mapping relationship trained using a multivariate regression method based on historical gait data from a large number of healthy individuals or the patient's unaffected side. Its input features include at least the aforementioned support duration, swing phase duration, pressure rise slope of each of the three regions, and peak occurrence time (a total of 8 input variables). The output is the estimated pressure values ​​for the medial forefoot, lateral forefoot, and heel center regions of the contralateral foot. The training principle of the mapping model is as follows: under standard bifoot acquisition conditions, the aforementioned parameters of both feet are recorded simultaneously. The parameters of the left foot are used as input, and the pressure values ​​of the corresponding regions of the right foot are used as output (or vice versa). The regression coefficient matrix is ​​solved using the least squares method to establish a mapping function from unilateral temporal and pressure dynamic characteristics to the spatial distribution of contralateral pressure.

[0142] For example, for an adult male of average build, when the pressure rise slope of the medial forefoot of his left foot during walking is 25 N / s and the peak occurs at 30% of the support phase, the mapping model estimates the pressure value of the corresponding area of ​​the right foot to be 210 N based on the average response pattern of the medial forefoot of the right foot in this type of gait pattern in the training samples.

[0143] After obtaining the estimated pressure values ​​for the three regions of the contralateral foot, the system possesses a set of measured pressure data (one foot) and a set of estimated pressure data (the contralateral foot). Based on the measured pressure data of the one foot, the system calculates the total weight-bearing capacity and pressure distribution on the medial and lateral sides of the forefoot. Then, following the calculation methods for the aforementioned weight-bearing balance score and weight-bearing capacity score, it determines the first rehabilitation efficacy index for that side. Simultaneously, based on the estimated pressure data of the contralateral foot (including estimated pressure values ​​for the three regions), the system uses the exact same calculation logic to determine the second rehabilitation efficacy index for the contralateral foot. It is important to emphasize that for the second rehabilitation efficacy index calculated based on the estimated data, the system will mark the estimated attribute in the background to ensure traceability for subsequent use.

[0144] Finally, the first rehabilitation efficacy index and the second rehabilitation efficacy index are substituted into the calculation formula for the left-right load symmetry score, i.e., K=1-|REI_measured-REI_estimated| / REI_max, where REI_max is the larger of the two, so as to obtain the left-right load symmetry score under the current target state.

[0145] To ensure the clinical reliability of the estimation results, this solution also includes a confidence assessment mechanism. During each estimation, the system calculates the prediction error distribution of the mapping model (based on the standard deviation of the prediction residuals of each sample in the training set) and the Mahalanobis distance between the current input feature vector and the model's training sample set. These two factors are combined to determine the confidence interval for this estimation. If the confidence interval is narrow (e.g., within ±5% of the predicted value), it is marked as high confidence, and the estimated data can be directly used in the calculation of the rehabilitation efficacy index. If the confidence interval exceeds a preset threshold, while calculating the symmetry score, explicit labeling information is added to the rehabilitation efficacy prompt, such as "The current symmetry score is based on unilateral estimation, and the confidence level is low. It is recommended to perform simultaneous bilateral foot sampling for verification on a later date." This helps doctors and patients interpret the assessment results reasonably and avoids misjudgments caused by estimation bias.

[0146] Through the above process, this solution can still provide a complete three-dimensional rehabilitation efficacy assessment with reference value even with only a single smart insole as a hardware constraint, significantly improving the system's clinical applicability and patient compliance.

[0147] Building upon the aforementioned embodiments, when calculating the rehabilitation efficacy index by weighted summation of the scores across the three dimensions, the initial weights of each score can be preset manually. However, addressing the "cold start" problem (i.e., insufficient historical data for effective reasoning by the strategy network) faced by reinforcement learning dynamic weight allocation strategies in the early stages of patient rehabilitation, this solution provides an initial weight determination method based on group similarity transfer. By mining group experience from historical rehabilitation data in the cloud, it generates personalized and clinically valuable initial weight combinations for new patients. The core idea of ​​this solution is that, under the same injury type, similar age, and similar baseline weight-bearing levels, the rehabilitation process and optimal weight allocation patterns of different patients share group commonalities. These commonalities can be extracted and transferred to new patients as effective prior knowledge before initializing their reinforcement learning strategy network.

[0148] For example, the individual characteristics data of patients are obtained, including injury type, rehabilitation stage, age, weight and baseline weight-bearing data; the individual characteristics data are matched with the group characteristics data of multiple historical rehabilitation patients in the cloud database, and the multiple historical patients with the highest similarity ranking are selected as reference groups; the historical weight sequence used by the reference groups at the same rehabilitation stage is extracted; and the initial weight of each score of the patient in the current assessment period is determined based on the historical weight sequence.

[0149] Specifically, when a new patient initiates their first rehabilitation assessment, the assessment terminal first acquires the patient's individual characteristic data. This individual characteristic data includes at least the type of injury (e.g., varus lateral ligament injury, valgus deltoid ligament injury, or high tibiofibular syndesmosis injury), the current rehabilitation stage (e.g., week post-surgery or early / mid / late rehabilitation), age, weight, and baseline weight-bearing data (i.e., the total plantar pressure values ​​during static standing and dynamic walking at the time of the initial assessment, serving as a benchmark for subsequent weight-bearing capacity calculations). This characteristic data can be obtained through manual input by the patient in the rehabilitation mini-program, configuration by the doctor through a cloud platform, or automatic baseline measurement during the initial insole collection. All of these are key discrete or continuous variables influencing the rehabilitation weight allocation strategy.

[0150] Subsequently, the assessment terminal encapsulates the aforementioned individual characteristic data into feature vectors, uploads them to the cloud-based rehabilitation data platform via a secure protocol, and calls the group similarity matching interface provided by the platform. The cloud database stores a large amount of desensitized group characteristic data from historical rehabilitation patients. Each historical patient's record includes their complete individual characteristic labels, daily or weekly R / K / Z score sequences, the actual weight combination sequences used in each assessment cycle, and the final rehabilitation results (such as whether the target was met, the time taken to meet the target, etc.). After receiving the feature vector of a new patient, the platform performs similarity matching calculations in the historical patient database. The similarity measure can use weighted Euclidean distance. For discrete features such as injury type and rehabilitation stage, one-hot encoding is first performed to convert them into numerical vectors, which are then combined with continuous features such as age and weight to form a complete feature space. Z-score standardization is performed on each dimension to eliminate dimensional differences. For example, the injury type is encoded as [1,0,0] for inversion, [0,1,0] for eversion, and [0,0,1] for high-level injury. The rehabilitation stage is normalized to a continuous value between 0 and 1 by the number of weeks.

[0151] After the similarity calculation is completed, the system sorts all historical patients from high to low similarity and selects the top-ranked historical patients as the reference group. The size of the reference group is preset to the top 50 (this number can be dynamically adjusted according to the amount of cloud data and the need for statistical robustness, but is generally no less than 30 cases to ensure the reliability of the group statistics).

[0152] After identifying the reference group, the system extracts the historical weight sequences used by that group at the same rehabilitation stage. The same rehabilitation stage refers to a time window that matches the current patient's stage. For example, if the current patient is in the 3rd week post-surgery, the system extracts the actual weight combination records for each patient in the reference group during the 3rd week (±3 days) post-surgery. Each historical patient may have weight records for one or more assessment cycles at the corresponding stage (e.g., daily assessment). The system statistically aggregates these weight records by dimension, calculating the reference group's average weight μ_R on the weight-bearing balance dimension, the average weight μ_K on the left-right weight-bearing symmetry dimension, and the average weight μ_Z on the weight-bearing capacity dimension. Simultaneously, the system calculates the standard deviations σ_R, σ_K, and σ_Z of the weights for each dimension to characterize the dispersion of weight allocation within the group at that stage.

[0153] Finally, the system determines the initial weights of each score component for the current patient in the current assessment period based on the average weights of the reference group. The simplest method is to directly use the group mean (μ_R, μ_K, μ_Z) as the initial weights. For example, if the average weights of the reference group at week 3 post-surgery are w_R=0.32, w_K=0.28, and w_Z=0.40, then these will be used as the initial weights for the weighted calculation during the first assessment of the new patient.

[0154] To further improve the adaptability of the initial weights to individual differences among new patients, the system can also adopt a weighted transfer strategy: assign different contribution weights to each historical patient in the reference group according to the similarity ranking, and assign higher contribution coefficients to patients with higher rankings (for example, the contribution coefficient of the ranked 1st is 1.0, and the contribution coefficient of the ranked 50th is reduced to 0.2, using linear or exponential decay). Then, a weighted average weight is calculated instead of a simple arithmetic average, so that the transfer results are closer to the actual selection of the most similar patients.

[0155] In addition, the system also outputs the group standard deviation of the weights of each dimension as a reference indicator of the group consensus. If the standard deviation is small (e.g., all less than 0.05), it indicates that the weight allocation of the group is highly consistent at this stage and the initial weights are highly reliable. If the standard deviation is large (e.g., more than 0.10), it indicates that there are large differences within the group. In this case, the initial weights are only for reference and the exploration range of subsequent reinforcement learning should be appropriately expanded.

[0156] Through the above process, this solution leverages the transfer of historical experience from a group in the cloud to provide new patients with an initial weight combination based on real rehabilitation trajectories, effectively addressing the policy void problem in the cold start phase of reinforcement learning strategies. This initial weight not only carries the successful experiences of similar populations at the same rehabilitation stage but also achieves preliminary personalization through individual feature matching. This provides ideal starting parameters for the subsequent training or online fine-tuning of the reinforcement learning strategy network, thereby shortening the strategy convergence time and enabling patients to obtain a clinically valuable rehabilitation efficacy index from their first assessment. As patients gradually accumulate their own assessment data (usually over 2 to 4 weeks), the system can progressively reduce the proportion of group transfer weights, shifting to primarily relying on the output of the strategy network trained on the patient's own data, achieving a smooth transition from group commonality guidance to individual-specific optimization.

[0157] Based on the above embodiments, in order to obtain the static rehabilitation efficacy index and the dynamic rehabilitation efficacy index, initial pressure data can be obtained. The initial pressure data is collected by the pressure sensors of each region over a preset time period. The initial pressure data includes static pressure data and dynamic pressure data. The static pressure data includes the average static pressure collected by the pressure sensors of each region over the preset time period. The dynamic pressure data includes the average pressure peak value of multiple gait cycles collected by the pressure sensors of each region over the preset time period. Then, the pressure data of the type required for the current assessment cycle can be selected from the initial pressure data.

[0158] The preset time period can refer to the period during which the data acquisition module uploads data, or it can refer to a data statistics window, such as 1 hour, 30 minutes, or 1 day. Within this time period, the data acquisition module can collect static pressure data, dynamic pressure data, or both.

[0159] For example, in a static state, if the pressure sensor on the medial side of the forefoot collects five pressure values ​​within one hour, the average of these five pressure values ​​can be used as the static pressure data for that hour. Similarly, the static pressure data from the other sensors for the same hour can be obtained. In a dynamic state, the three sensors collect pressure peak values ​​from multiple gait cycles within that hour, and the average of these pressure peak values ​​is then used as the dynamic pressure data.

[0160] Initial stress data refers to the raw summary data packets uploaded by the data acquisition module every hour (with a preset time period of 1 hour). It may include the static average stress value for each region (i.e., static stress data), the dynamic peak average stress value for each region (i.e., dynamic stress data), and the total number of steps within the hour. Different fields can be used to distinguish stress data in different states when it is packaged and sent. To facilitate rapid data retrieval, the initial stress data may also include a status flag, which can be a one-byte identifier field indicating the data type within the preset time period, such as static, dynamic, or mixed types.

[0161] After acquiring initial stress data, the assessment terminal can determine the data type required for this assessment based on the current rehabilitation stage or user configuration. For example, if the patient is in the early stage of rehabilitation and the total number of steps is less than the set number, the required data type is static; if the patient is in the middle or late stage of rehabilitation and the total number of steps is greater than the set number, the required data type is dynamic. If the system is configured for dual-state synchronous assessment, the required data type is mixed.

[0162] After obtaining the required data type, the evaluation terminal can first check the status flags in the initial pressure data to determine whether the initial pressure data meets the requirements. For example, if the required data type is static, it checks whether the status flags represent a static type or a mixed type. If they do, static pressure data is extracted from it and used as input data for calculating the three scores. If they do not meet the requirements, the initial pressure data is discarded, and calculations are performed only after new pressure data is received.

[0163] In the above implementation process, by obtaining the average static pressure and the average dynamic pressure peak within a preset time period, and flexibly selecting the data type required for the current assessment cycle, the rehabilitation efficacy assessment can be adapted to the monitoring needs of different rehabilitation stages.

[0164] Based on the above embodiments, after obtaining the rehabilitation efficacy index, the assessment terminal can also generate rehabilitation efficacy prompts, which include at least one of the following: a trend chart of the rehabilitation efficacy index changing over time, a rehabilitation progress report, rehabilitation training guidance suggestions, and abnormal warning information. Additionally, it can generate and display at least one of the following: a pressure heat map of each area, a pressure distribution percentage curve, and physician rehabilitation assessment suggestions. All of this information can be displayed through the aforementioned rehabilitation mini-program.

[0165] A pressure thermogram is a visual representation of pressure distribution in three areas of the foot: the medial forefoot, the lateral forefoot, and the heel. It typically uses a gradient from green (low pressure) to red (high pressure) and converts absolute pressure values ​​into a percentage of the patient's body weight for cross-sectional comparisons between patients of different weights.

[0166] When generating a pressure heatmap, after receiving the latest pressure data (static average or dynamic peak average) uploaded by the insole, the assessment terminal first extracts the absolute pressure values ​​of the three regions, and then calls the weight normalization module: if the patient has entered their weight (in kilograms), the pressure value of each region is divided by (weight × 9.8) to obtain the percentage of body weight in that region; if no weight has been entered, the absolute pressure is used directly, but labeled "absolute pressure" in text. Next, according to a preset color mapping table (for example, in static conditions, less than 16% pressure corresponds to green, 16%-20% to yellow, and more than 20% to red; in dynamic conditions, less than or equal to 40% corresponds to green, and greater than 46% to red), the corresponding RGB color value is calculated for each region. The assessment terminal uses the relevant software of the rehabilitation mini-program to draw a simplified outline of the foot, fills the three regions with the calculated colors, and displays the pressure value (rounded to an integer) or percentage in the center of each region. The patient can click the "switch" button on the heatmap to select whether to display the pressure distribution when standing statically or the dynamic peak distribution during the most recent walking session. In addition, if a patient wears both insoles simultaneously, the app will display pressure heat maps of both feet side-by-side, indicating the degree of asymmetry with arrows or text. The pressure heat map refreshes automatically whenever new data is received, and users can also manually scroll the timeline to view historical heat maps.

[0167] A pressure distribution curve is a line graph that displays the percentage of pressure in each area relative to the total pressure. It typically uses hours as the horizontal axis and percentage as the vertical axis to observe changes in load balance over time. For example, the assessment terminal can query the database for effective pressure data for each hour over the past 24 hours (or a user-defined time period) (preferably using the dynamic peak average, or the static average if unavailable). The percentage for each area is calculated hourly, and then three lines are plotted (medial forefoot, lateral forefoot, and heel), with time on the horizontal axis and percentage (0-100%) on the vertical axis. Users can click the legend to hide / show specific curves.

[0168] The rehabilitation progress report can be an automatically generated summary of text and charts on a daily, weekly, or monthly basis, including the REI mean, sub-indicator scores, steps, wearing time, and comparison with the baseline.

[0169] Abnormal warning information can be alarm texts triggered by real-time stress data or historical indicators (such as continuous low balance, excessive load peak, abnormal rehabilitation efficacy index), and pushed through pop-ups or service notifications.

[0170] The trend chart is a line chart or bar chart that shows the changes of the Rehabilitation Efficacy Index (REI) and its sub-indicators (weight-bearing balance score, left-right weight-bearing symmetry score, and weight-bearing capacity score) over time (hours, days, weeks). The horizontal axis represents time, and the vertical axis represents the score value from 0 to 1. Target threshold lines or rehabilitation target lines set by doctors can be superimposed.

[0171] When generating the trend chart, the assessment terminal can query the daily REI values ​​for the past few days (e.g., the last 30 days) from the local database or cloud storage (if hourly REIs are available, they are aggregated into a daily average REI). A line chart is drawn using the mini-program's native chart component: the horizontal axis represents the date, and the vertical axis represents the REI value (range 0-1), with the following auxiliary elements added: a rehabilitation target line (e.g., a dashed line representing the doctor's discharge standard REI ≥ 0.85); milestone markers (a medal icon is displayed on the line when the REI first reaches 0.3, 0.5, 0.7, and 0.9); and a sliding information window (users click or long-press a day to display the specific REI value for that day and the scores of the three sub-indicators). In addition, the trend chart page provides a "Sub-Indicator Decomposition" toggle button, allowing users to view the change curves of the weight-bearing balance score, left-right weight-bearing symmetry score, and weight-bearing capacity score separately, facilitating the identification of weak points in rehabilitation. Every Monday morning, the system can automatically generate a trend report, which will include a screenshot of the REI curve for the past 7 days and a brief text interpretation (e.g., "This week's average REI is 0.71, an increase of 0.08 from last week, indicating significant improvement in weight-bearing capacity"), and provide it to patients in image or PDF format for them to save or share with their doctors.

[0172] Rehabilitation training guidance suggestions can be personalized rehabilitation suggestions automatically generated by the rule engine or lightweight AI model of the assessment terminal based on the current REI value and historical trend. These suggestions include training type, daily training duration, maximum load, and precautions.

[0173] Specifically, the assessment terminal has a built-in rule-tree-based decision logic that automatically generates text-based rehabilitation suggestions based on the current REI value, rehabilitation stage (number of days), and recent trends. The generation steps are as follows: Determining the recovery stage: If REI < 0.4, early stage; 0.4 ≤ REI < 0.7, middle stage; REI ≥ 0.7, late stage.

[0174] Incorporating trend correction: If the REI growth rate in the most recent week is <0.05 and the REI is still below 0.7, add "Rehabilitation progress is slow, you can appropriately increase the training intensity or contact your doctor to adjust the plan" to the recommendations; if the REI drops by more than 0.1, add "Rehabilitation has regressed, please reduce the load and consult your doctor".

[0175] Output specific suggestion template: Early stage: "You are currently in the early stage of rehabilitation, and it is recommended to focus on non-weight-bearing training. Perform ankle pump exercises (pointing and flexing your toes) 3 sets a day, 20 repetitions per set; static standing training should not exceed 10 minutes per day in total, and pay attention to even weight distribution on both feet." Mid-term: "You have entered the partial weight-bearing stage. It is recommended to walk for a total of 30 minutes every day, with the affected foot bearing no more than 50% of your body weight; when walking, pay attention to landing on the entire sole of your foot and avoid landing only on your heel; combine this with calf raises, 2 sets of 15 repetitions each day." Later: "Your dynamic function has recovered well and you can gradually return to normal walking. Please pay special attention to gait symmetry: keep your left and right strides consistent when walking, and you can check yourself by looking in a mirror or recording yourself with your mobile phone; add 5 minutes of balance training (standing on one leg) every day." Personalized suggestions: If the weight-bearing capacity score is below 0.5, the suggestion will include "Weak weight-bearing capacity, please do more static standing endurance training"; if the balance score is below 0.6, the suggestion will include "Uneven pressure on the inner and outer sides of the forefoot, consciously shift the center of gravity from the outer side to the inner side when walking"; if the symmetry score is below 0.7 (when wearing the device on both feet), the suggestion will include "Uneven recovery progress between the left and right legs, try to strengthen the support on the affected side when walking".

[0176] The generated suggested text is displayed on the "Today's Guidance" card in the mini-program, along with a short animated GIF or video link (such as an ankle pump exercise demonstration). Patients can click the "Done" button to record compliance, and the system accumulates points that can be redeemed for achievement badges.

[0177] Doctor rehabilitation assessment suggestions refer to text suggestions manually entered by registered doctors after viewing patient data through the cloud-based rehabilitation data platform (such as "strengthen weight-bearing on the inner side of the forefoot and do 10 minutes of heel raises daily"). These suggestions are synchronized to the assessment terminal via HTTPS and displayed on a specific page of the mini-program.

[0178] For example, the assessment terminal proactively requests the cloud-based rehabilitation data platform every 4 hours or upon startup to check if there are any new doctor recommendations for the current patient. The data returned from the cloud includes fields such as recommendation ID, content, doctor's name, publication time, and an emergency flag. Upon receiving this data, the mini-program stores the recommendation locally and displays a red dot at the top of the homepage (if any recommendations are unread). On the "Doctor Recommendations" page, all historical recommendations are displayed in reverse chronological order. Each recommendation is presented as a card, including the doctor's avatar, department, recommendation content, and publication date. For recommendations marked "Emergency," the card background color changes to light red, a modal pop-up alert appears, and the mini-program's voice broadcast function reads the recommendation content aloud. After reading, the patient can click the "Read" button, and the mini-program sends the status back to the cloud for the doctor to confirm that the patient has been informed. In addition, the doctor recommendations page provides a "Feedback" input box where patients can enter text feedback on the implementation status (e.g., "External weight-bearing training has been added as recommended"). The feedback is synchronized to the cloud for remote monitoring by the doctor.

[0179] By generating and displaying the aforementioned information, the assessment terminal transforms the abstract rehabilitation efficacy index into an intuitive and operable form, which not only enhances patients' sense of participation in self-management but also provides an effective bridge for remote communication between doctors and patients.

[0180] Based on the above embodiments, after obtaining various pressure data and scores, anomaly warnings can be issued based on these data.

[0181] For example, after receiving dynamic pressure data uploaded by the insole each time, the assessment terminal extracts the single maximum peak value (i.e., the highest pressure value in a certain area within a single gait cycle) from all dynamic data collected within that hour. This maximum peak value is compared with a safety threshold preset by the doctor. The safety threshold can be set individually by the doctor for each patient through a cloud platform (e.g., "the total dynamic peak value of the affected foot should not exceed 40% of body weight"). If the maximum peak value exceeds 1.2 times the threshold (or exceeds the threshold for two consecutive hours), an overload warning is triggered. The warning information is sent via application service notification or a pop-up window in a mini-program.

[0182] For example, the assessment terminal can calculate the load balance score for the past 24 hours every morning. If the average load balance score for the day is below 0.6, and the average load balance score for the previous two days is also below 0.6 (i.e., imbalance for 3 consecutive days), a load imbalance warning will be triggered.

[0183] For example, the assessment terminal calculates the average left-right weight-bearing symmetry score for the past week every Monday and compares it with the average score of the previous week. If the symmetry score for this week decreases by more than 0.2 compared to last week, and the absolute value of the symmetry score for this week is below 0.7, a symmetry deterioration warning is triggered. At the same time, the system will display the changes in REI on both sides separately to help determine whether the deterioration mainly originates from the affected side or the healthy side.

[0184] For example, the assessment terminal can track the effective duration of patients wearing insoles daily. Effective duration is defined as the total number of hours with at least one effective pressure data point (static or dynamic) per hour within a day. If the effective duration is less than 8 hours in a day, it is recorded as insufficient compliance. When insufficient compliance occurs for 5 consecutive days, a low compliance warning is triggered.

[0185] For example, to avoid frequently disturbing patients, the system has a cooling-off period for the same type of alert: after a single alert is pushed out, the same type of alert will not be pushed out again within 24 hours. If an abnormal state persists for more than 7 days and the patient has not taken effective measures (e.g., the balance score is still below 0.6), the system will send an escalation alert to the doctor's end and display a "high-risk" icon in the mini-program, suggesting that the doctor proactively contact the patient by phone. All alert events are recorded locally and in the cloud, forming an alert log for doctors to review and analyze.

[0186] Through the aforementioned abnormal early warning mechanism, the assessment terminal can monitor risk factors in the rehabilitation process in real time and guide patients to adjust their behavior in a timely manner, thereby effectively reducing the probability of secondary injury and improving rehabilitation safety.

[0187] Please refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of an assessment terminal for performing an ankle joint rehabilitation efficacy assessment method, provided in an embodiment of this application. The assessment terminal may include: at least one processor 310, such as a CPU, at least one communication interface 320, at least one memory 330, and at least one communication bus 340. The communication bus 340 is used to establish communication between these components. In this embodiment, the communication interface 320 is used for signaling or data communication with other node devices. The memory 330 may be a high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 330 may also be at least one storage device located remotely from the aforementioned processor. The memory 330 stores computer-readable instructions; when these computer-readable instructions are executed by the processor 310, the assessment terminal performs the aforementioned method process.

[0188] Understandable. Figure 3 The structure shown is for illustrative purposes only; the evaluation terminal may also include a comparison... Figure 3 The more or fewer components shown, or having the same Figure 3 The different configurations shown. Figure 3 The components shown can be implemented using hardware, software, or a combination thereof.

[0189] This application provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it performs the method process executed by the evaluation terminal in the above method embodiments.

[0190] This embodiment discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can perform the methods provided in the above-described method embodiments.

[0191] In summary, this application provides a method, system, assessment terminal, and storage medium for assessing ankle rehabilitation efficacy. By modeling the weight allocation problem as a sequential decision model, using the current scores and trends of each dimension as state input, and maximizing the cumulative improvement of the rehabilitation efficacy index over a future period as the optimization objective, the strategy network trained by the strategy gradient algorithm can intelligently adjust the weight allocation according to the patient's current rehabilitation status. This allows the weights of the three dimensions—weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity—to adaptively adjust based on the patient's current rehabilitation state and its trends, achieving dynamic alignment between the assessment focus and the patient's actual weaknesses. Furthermore, by using the cumulative improvement of the future rehabilitation efficacy index as the optimization objective, the weight allocation is no longer limited to fitting the current state but actively guides the patient towards the long-term optimal rehabilitation trajectory. This effectively solves the intervention decision-making problem during the rehabilitation plateau period and significantly improves the accuracy of individualized assessment and the foresight of rehabilitation training guidance. The rehabilitation efficacy index is determined by comprehensively considering three dimensions: weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity. This provides a multi-dimensional quantitative assessment method for ankle rehabilitation, ultimately generating intuitive rehabilitation efficacy prompts. This achieves objective quantification and visual feedback on the ankle rehabilitation process, providing patients with clear guidance on their rehabilitation progress. At the same time, it provides reliable data for doctors to develop and adjust rehabilitation plans and for patients to manage their own rehabilitation.

[0192] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0193] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0194] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0195] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0196] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for evaluating the effectiveness of ankle joint rehabilitation, characterized in that, The method includes: Acquire pressure data, which includes static pressure data and / or dynamic pressure data. The pressure data is collected by pressure sensors installed in various areas of the insole, and each area corresponds to the medial forefoot, lateral forefoot, and center heel position of the patient's foot. Based on the pressure data, the patient's weight-bearing balance score, left-right weight-bearing symmetry score, and weight-bearing capacity score are determined under the target state. The weight-bearing balance score is determined based on the difference in pressure data collected by pressure sensors at the medial and lateral positions of the forefoot under the target state. The left-right weight-bearing symmetry score is determined based on the difference in rehabilitation efficacy index between the patient's left foot and right foot under the target state. The weight-bearing capacity score is determined based on the patient's current weight-bearing capacity and baseline weight-bearing capacity. The target state includes static and / or dynamic states. The weight-bearing balance score, the left-right weight-bearing symmetry score, and the weight-bearing capacity score are weighted and calculated to obtain the rehabilitation efficacy index. The weight of each score is determined by a dynamic weight allocation strategy. The dynamic weight allocation strategy includes: constructing the patient's rehabilitation process as a sequential decision model, using the scores and trends of weight-bearing balance, left-right weight-bearing symmetry, and weight-bearing capacity in the current assessment period as the state space, the weight combination corresponding to the three dimensions as the action space, and the cumulative increase in the rehabilitation efficacy index in a preset future time period as the cumulative reward; training the sequential decision model through a strategy gradient algorithm, and outputting the optimal weight combination in the current state to maximize the cumulative reward. Rehabilitation efficacy prompts are generated based on the rehabilitation efficacy index.

2. The method according to claim 1, characterized in that, The initial weights for each score were obtained in the following way: The individual characteristic data of the patient are obtained, including injury type, rehabilitation stage, age, weight, and baseline weight-bearing data; The individual characteristic data is matched with the group characteristic data of multiple historical rehabilitation patients in the cloud database, and the multiple historical patients with the highest similarity ranking are selected as reference groups. Extract the historical weight sequence used by the reference group at the same rehabilitation stage; The initial weights for each score of the patient in the current assessment period are determined based on the historical weight sequence.

3. The method according to claim 1, characterized in that, The rehabilitation efficacy index of the patient's left foot and the rehabilitation efficacy index of the patient's right foot were determined in the following manner: When only pressure data of one foot of the patient is obtained, gait timing parameters and pressure dynamic parameters in the target state are extracted from the pressure data; Based on the gait timing parameters and the pressure dynamic parameters, the estimated pressure data of the patient's contralateral foot under the corresponding target state is estimated using a preset mapping model; A first rehabilitation efficacy index is determined based on the pressure data of the unilateral foot, and a second rehabilitation efficacy index is determined based on the estimated pressure data of the contralateral foot.

4. The method according to claim 1, characterized in that, The acquisition of stress data includes: Acquire initial pressure data, which is collected by pressure sensors in each region over a preset time period. The initial pressure data includes: static pressure data and dynamic pressure data. The static pressure data includes the average static pressure collected by pressure sensors in each region over the preset time period. The dynamic pressure data includes the average pressure peak values ​​of multiple gait cycles collected by pressure sensors in each region over the preset time period. Select the required data type of stress data for the current evaluation cycle from the initial stress data.

5. An ankle joint rehabilitation efficacy assessment system, characterized in that, The system includes: The insole is equipped with pressure sensors in various areas, which correspond to the medial forefoot, lateral forefoot, and heel center of the patient's foot. The pressure sensors are used to collect pressure data. The insole is also equipped with posture sensors, which are used to collect the patient's posture data. An evaluation terminal, the evaluation terminal being used to perform the method according to any one of claims 1-4.

6. The system according to claim 5, characterized in that, The insole is also equipped with a data acquisition module, which communicates with each pressure sensor and the posture sensor; The data acquisition module is used to acquire data using an interrupt mechanism that employs timer interrupts and acceleration threshold interrupts. The data acquisition module is used to check whether static pressure data has been acquired within the current time window under the timed interrupt mechanism. If it has not been acquired and both the pressure data and the attitude data meet the static conditions, it wakes up each pressure sensor to acquire static pressure data and enters sleep mode after the acquisition is completed. The data acquisition module is used to wake up each pressure sensor to collect dynamic pressure data under the acceleration threshold interruption mechanism, record the pressure peak value and its average value during the gait cycle, and enter sleep mode after the collection is completed.

7. The system according to claim 5, characterized in that, The insole is also provided with a communication module, which communicates with the evaluation terminal, and / or, the insole is also provided with a lithium battery power supply module, which is used to power the various devices in the insole, and / or, the insole is also provided with an offline storage module, which is used to store the collected pressure data and attitude data; And / or, the assessment terminal is equipped with a rehabilitation mini-program; The rehabilitation mini-program is used to generate and display a pressure heatmap based on the acquired pressure data. And / or, the rehabilitation mini-program is also used to generate and display a trend chart based on the rehabilitation efficacy index; And / or, the rehabilitation mini-program is also used to generate and display rehabilitation training guidance suggestions based on the rehabilitation efficacy index; And / or, the rehabilitation mini-program is also used to display the doctor's rehabilitation assessment suggestions; And / or, the system further includes a cloud-based rehabilitation data platform for storing the patient's historical rehabilitation data for doctors to remotely view and issue rehabilitation assessment suggestions, the cloud-based rehabilitation data platform communicating with the assessment terminal.

8. An evaluation terminal, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions that, when executed by the processor, perform the method as described in any one of claims 1-4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the method as described in any one of claims 1-4.

10. A computer program product, characterized in that, It includes computer program instructions, which, when read and executed by a processor, perform the method as described in any one of claims 1-4.

Images